Logic via Algebra

Sam Chong Tay

A Senior Exercise in Mathematics
Kenyon College
November 29, 2012
 Abstract

The purpose of this paper is to gain insight to mathematical logic through

an algebraic perspective. Although readers with a working knowledge of
mathematical logic will see very familiar concepts, the only prerequisite to
this paper is an undergraduate level course in abstract algebra. We begin
by defining a propositional language and proving a few simple theorems,
to be followed by an introduction to boolean algebra and its consistency
with the propositional calculus. We will then show how common notions in
propositional logic, such as the soundness and completeness theorems, can
be viewed as algebraic notions.
Contents
Preface 2

1 Introduction to Propositional Calculus 3

1.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Deductions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Semantic Interpretations . . . . . . . . . . . . . . . . . . . . . 5
1.4 Important Properties of L . . . . . . . . . . . . . . . . . . . . 6

2 Introduction to Boolean Algebra 7

2.1 Defining a Boolean Algebra . . . . . . . . . . . . . . . . . . . 7
2.2 Properties of a Boolean algebra . . . . . . . . . . . . . . . . . 10
2.3 Examples of Boolean Algebras . . . . . . . . . . . . . . . . . 14

3 Structure of a Boolean Algebra 19

3.1 Homomorphisms . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 Ideals and Filters . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.3 Ultrafilters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

4 Applying Algebra 29
4.1 Making Connections . . . . . . . . . . . . . . . . . . . . . . . 30
4.2 Soundness, Completeness, and Compactness . . . . . . . . . . 36

5 Conclusion 39

1
Preface
Students of mathematics learn early on the rules of logic by which we can
write rigorous proofs. It is quite arguable that the axioms and rules of in-
ference of a particular mathematical theory are the most important parts,
as they constitute the foundation of the theory. As mathematicians are
wont to do, we ask ourselves how these concepts generalize – that is, given
an axiomatic system and rules of inference, aside from particular theorems
that can be proven within the system, what can we say about the system
itself? The question is metamathematical, and it is quite an important
exploration. Previous work in the 20th century, such as Gödel’s Incom-
pleteness Theorems, tells us that there are certain limitations on different
axiomatic systems. In particular those theorems tell us that there is no
complete axiomatic system describing simple arithmetic of natural num-
bers, which means there will be statements that are not provably true or
false. However, as we will show in this paper, the propositional calculus is
complete.
The final results of this paper are theorems about the propositional cal-
culus, and while they are important results, they can be proven directly from
the definitions given in Section 1 without much difficulty. It might take an
undergraduate level class all of two weeks to build up enough theory to prove
these results. However, the importance of this paper does not lie in the the-
orems themselves, but in their proofs. After introducing Boolean algebra in
Section 2 and building algebraic machinery in Section 3, the common the-
orems of propositional calculus in Section 4 can be proven by viewing the
propositional calculus algebraically. The connections are very enlightening,
and quite elegant as well.
(A note on methodology: at the beginning of each section, I describe
from which source the definitions and theorems may be found. For the most
part, the proofs are my own work, as many of the theorems are stated as
problems in [1]. The few proofs that rely on arguments from my sources are
cited accordingly. The theorems in Section 4, as already stated, are common
in the study of mathematical logic. The algebraic proofs however, are solely
my own. I worked backwards to find the algebraic results needed, and wrote
Section 3 accordingly.)

2
1 Introduction to Propositional Calculus
The propositional calculus is very formally the study of particular strings of
letters, made up of a certain alphabet. To build up such a study, there are
copious necessary definitions. The definitions used in this first section are
from [2]. For those who are familiar with these concepts, the first section
may be skipped or perhaps skimmed, to become familiar with the notation
used throughout this paper. After Section 1, we assume a basic knowledge
of set theoretic notions.

1.1 Definitions
Let L denote the propositional language, which consists of an alphabet, from
which we can construct sentences. The alphabet has six letters, namely

¬, ⇒, p, +, (, ).

Definition 1.1.1. The first two letters ¬ and ⇒, which we pronounce as

“not” and “implies” respectively, are propositional connectives.

Definition 1.1.2. The letter p followed by any finite number of occurrences

of +, such as
p , p+ , p+++++++ ,
is a propositional variable.

Definition 1.1.3. A sentence of L is defined inductively as follows:

(i) The propositional variables p , p+ , p++ , . . . are sentences.

(ii) If x is a sentence, then so is (¬x).

(iii) If x and y are sentences, then so is (x ⇒ y).

The sentence (¬x) is called the negation of x, and the sentence (x ⇒ y) is

called the implication of the sentence y from x.

With these very formal definitions in place we can unambiguously in-

troduce some abbreviations. The first natural abbreviation is to denote
propositional variables p , p+ , and p++ , as English letters such as p, q, and
r. We also introduce the new connective symbols ∨ and ∧, defined so that

(x ∨ y) = ((¬x) ⇒ y)

3
and
(x ∧ y) = (¬(x ⇒ (¬y))).
These symbols are pronounced “or” and “and” respectively. From here we
also introduce the connective symbol ⇔ pronounced “if and only if” and
defined as
(x ⇔ y) = ((x ⇒ y) ∧ (y ⇒ x)).
Of course these new connective symbols could have been included in the for-
mal alphabet, but many formal definitions, such as those of sentences and
deductions, are much more concise when only needing to deal with two for-
mal connectives. Lastly, as with many areas in mathematics, we will allow
the omission of parentheses in scenarios that lead to no ambiguities. One
example is the outermost parentheses of (x ⇒ y), which are not necessary in
describing the sentence itself. As with typical conventions, we allow prece-
dence of negation over all other operations. Also among binary operations,
∨ and ∧ take precedence over ⇒ and ⇔. For example in the definition of ∧
above, we have
x ∧ y = ¬(x ⇒ ¬y).

1.2 Deductions
There is a certain type of sentence of L called a theorem. We will describe
these in full shortly, but first we describe a particular kind of theorem called
an axiom. If x, y, and z are sentences, then each of the sentences

A1 x∨x⇒x
A2 x⇒x∨y
A3 x∨y ⇒y∨x
A4 (x ⇒ y) ⇒ (z ∨ x ⇒ z ∨ y)

are axioms, and all axioms are one of these.

In order to completely describe the set of theorems, we specify a “rule
of inference” called modus ponens. This rule states that whenever x and y
are sentences such that x and x ⇒ y are theorems, then y is also a theorem.
Definition 1.2.1. We say that y is a theorem and write
`y
if and only if there is a finite sequence of sentences, whose last term is y,
such that each sentence in the sequence is either (i) an axiom, or (ii) the

4
result of applying modus ponens to some pair of preceding sentences in the
sequence. The finite sequence just described is a deduction of y.
Example 1.2.2. The following is a deduction of the theorem x ⇒ x as
in the previous definition, where the justifications in parentheses are not
actually part of the sequence.

x⇒x∨x (A2)
x∨x⇒x (A1)
(x ∨ x ⇒ x) ⇒ ((x ⇒ x ∨ x) ⇒ (x ⇒ x)) (A4)
(x ⇒ x ∨ x) ⇒ (x ⇒ x) (MP 2,3)
x⇒x (MP 1,4)

Definition 1.2.3. Let A be a set of sentences of L. We say that a sentence

y is deducible from A and write
A`y
if and only if there is a finite sequence of sentences, whose last term is y,
such that each sentence in the sequence is either (i) an axiom, (ii) in A, or
(iii) the result of applying modus ponens to some pair of preceding sentences
in the sequence. The finite sequence just described is a deduction of y from
A.
Definition 1.2.4. A set of sentences A of L is said to be inconsistent if
A ` x ∧ ¬x for some sentence x. A set of sentences is consistent if it is not
inconsistent.

1.3 Semantic Interpretations

All of the definitions thus far are purely syntactical, built mechanically from
certain strings of symbols of L. Next we have semantic definitions.
Definition 1.3.1. A truth valuation for L is a particular mapping v as-
signing each sentence of L to a member of the set {0, 1}. We define such
a truth valuation inductively. First let v0 be any function from the set of
propositional variables into the set {0, 1}. Then we extend v0 uniquely into
a function v with a domain that is the set of all sentences of L, such that
for all sentences x and y,

v(¬x) = 1 if and only if v(x) = 0

v(x ⇒ y) = 1 if and only if v(x) = 0 or v(y) = 1.

5
If these conditions are met, then v is a truth valuation.
We can think of these two numbers as truth values, where 0 is false and
1 is true.
Definition 1.3.2. If x is a sentence such that v(x) = 1 for all possible truth
valuations v, then we call x a tautology and write  x.
Definition 1.3.3. If y is a sentence such that v(y) = 0 for all possible truth
valuations v, then we call y a contradiction.
It is immediate from the definition of truth valuation that if y is a con-
tradiction, then ¬y is a tautology. Hence, in this case we write  ¬y.
Definition 1.3.4. A set of sentences S of L is satisfiable if there is some
truth valuation v such that v(x) = 1 for all x ∈ S. In this case we say v
satisfies S.
Definition 1.3.5. Let y be a sentence and S be a set of sentences of L. If
for all truth valuations v satisfying S, that is for all v such that v[S] = {1},
it is also the case that v(y) = 1, then we say that y is a semantic consequence
of S and write S  y.
We have now set the stage completely for propositional logic. Of course,
what we would like is for the semantic and syntactic concepts to coincide;
this is the whole point of defining such a language, to put our intuitive logical
reasoning (semantic) into a consistent, purely mechanical form (syntactic).
We will consider these questions through algebraic means later on, but first
we need to make note of a few properties of the propositional calculus.

1.4 Important Properties of L

We end this section with the following theorems of L.

T1 x∨y ⇔y∨x T2 x∧y ⇔y∧x

T3 x ∨ (y ∨ z) ⇔ (x ∨ y) ∨ z T4 x ∧ (y ∧ z) ⇔ (x ∧ y) ∧ z
T5 (x ∨ y) ∧ y ⇔ y T6 (x ∧ y) ∨ y ⇔ y
T7 (x ∨ y) ∧ z ⇔ (x ∧ z) ∨ (y ∧ z) T8 (x ∧ y) ∨ z ⇔ (x ∨ z) ∧ (y ∨ z)
T9 x ∧ (y ∨ ¬y) ⇔ x T 10 x ∨ (y ∧ ¬y) ⇔ x

There is really nothing deep in formal deductions, and for the purpose
of this paper they are more cumbersome than enlightening. Thus we omit
them here and refer the skeptics to [2].

6
2 Introduction to Boolean Algebra
The following section will deal with Boolean algebra in a very general setting,
with definitions, examples, and propositions from [1, Ch. 4 §1-2]. It is a
lengthy excursion from the propositional calculus, but we will see many
familiar notions from Section 1 disguised in algebraic notation.

2.1 Defining a Boolean Algebra

Definition 2.1.1. A lattice is a non-empty partially ordered set hL, ≤i in
which each x, y ∈ L has a supremum (join) x ∨ y and an infimum (meet)
x ∧ y.

An obvious characterization that will be used frequently is the fact that

x≤y iff x∨y =y iff x ∧ y = x.

As it turns out, Boolean algebra was first defined with sets as elements, so
the following example has historical significance.

Example 2.1.2. Let Y be a set. We have the power set under inclusion
hP(Y ), ⊆i as a lattice, where

A∨B =A∪B and A ∧ B = A ∩ B.

It should be clear that this is a lattice; it is certainly obvious from set

theoretic facts that ⊆ is a partial ordering on P(Y ). To see that A ∪ B is in
fact the supremum of A and B, note that of course A ⊆ A∪B and B ⊆ A∪B.
If A ⊆ C and B ⊆ C, it follows that A ∪ B ⊆ C. Hence A ∨ B = A ∪ B.
The meet of A and B is similarly seen to be their intersection.

A first glimpse at the connection to propositional calculus is seen in the

following proposition.

Proposition 2.1.3. The following properties hold for any elements x, y, z

of a lattice L:

B1 x∨y =y∨x x∧y =y∧x

B2 x ∨ (y ∨ z) = (x ∨ y) ∨ z x ∧ (y ∧ z) = (x ∧ y) ∧ z
B3 (x ∧ y) ∨ y = y (x ∨ y) ∧ y = y

7
Proof. B1 is rather obvious when we consider

x ∨ y = sup{x, y} = sup{y, x} = y ∨ x,

and similarly

x ∧ y = inf{x, y} = inf{y, x} = y ∧ x.

For B2 we have
x ≤ x ∨ (y ∨ z)
and
(y ∨ z) ≤ x ∨ (y ∨ z),
from which we also have

y ≤ x ∨ (y ∨ z) and z ≤ x ∨ (y ∨ z).

The fact that x ≤ x ∨ (y ∨ z) and y ≤ x ∨ (y ∨ z) together imply that

(x ∨ y) ≤ x ∨ (y ∨ z).

In the same way, since z ≤ x ∨ (y ∨ z) we have

(x ∨ y) ∨ z ≤ x ∨ (y ∨ z).

An identical argument shows that x ∨ (y ∨ z) ≤ (x ∨ y) ∨ z, and we conclude

x ∨ (y ∨ z) = (x ∨ y) ∨ z. The proof for x ∧ (y ∧ z) = (x ∧ y) ∧ z is very
similar. For B3, we know immediately that

y ≤ y ∨ (x ∧ y).

We also have
y≤y and (x ∧ y) ≤ y,
which together imply that y ∨ (x ∧ y) ≤ y. Hence by antisymmetry we
conclude (x ∧ y) ∨ y = y. For the counterpart of B3, we have immediately
that
(x ∨ y) ∧ y ≤ y,
and of course
y≤y and y ≤ (x ∨ y).
The two previous inequalities imply that

y ≤ (x ∨ y) ∧ y,

and again by antisymmetry we conclude (x ∨ y) ∧ y = y.

8
Definition 2.1.4. A lattice L is distributive if for all x, y, z ∈ L,

x ∧ (y ∨ z) = (x ∧ y) ∨ (x ∧ z),

x ∨ (y ∧ z) = (x ∨ y) ∧ (x ∨ z).
Curiously, each of the two distributive properties above implies the
other [1]. To show this, suppose the first condition holds. Then

(x ∨ y) ∧ (x ∨ z) = [(x ∨ y) ∧ x] ∨ [(x ∨ y) ∧ z]
= [(x ∨ y) ∧ x] ∨ [(x ∧ z) ∨ (y ∧ z)]
= x ∨ [(x ∧ z) ∨ (y ∧ z)]
= [x ∨ (x ∧ z)] ∨ (y ∧ z)
= x ∨ (y ∧ z).

The first three equalities follow from the first distributive property and the
last two follow from Proposition 2.1.3. The converse is proved similarly.
Definition 2.1.5. A lattice L is complemented if there exist both least and
greatest elements in L, denoted 0 and 1 respectively, and if for each x ∈ L
there exists y ∈ L such that

x∨y =1 and x ∧ y = 0.

Proposition 2.1.6. The only complemented lattice that is totally ordered

is trivial [1].
Proof. Let L be a totally ordered and complemented lattice, and let x ∈ L.
Then x has a complement y such that x ∨ y = 1 and x ∧ y = 0. We know
that either x ≤ y or y ≤ x. In the first case, if x ≤ y we must have x ∨ y = y
and x ∧ y = x, which of course means that y = 1 and x = 0. If instead y ≤ x
then x ∨ y = x and x ∧ y = y, so that x = 1 and y = 0. Hence in any case
we have L = {0, 1}.

Proposition 2.1.7. An element of a distribute lattice has at most one

complement.
Proof. Let L be a distributive lattice with a 0 and a 1, and suppose y and
y 0 are both complements of an element x ∈ L. Then we have

x ∨ y = 1, x ∨ y 0 = 1,
x ∧ y = 0, x ∧ y 0 = 0,

9
and

y = y ∨ 0 = y ∨ (x ∧ y 0 ) = (y ∨ x) ∧ (y ∨ y 0 )
= 1 ∧ (y ∨ y 0 )
= y ∨ y0,

so that y 0 ≤ y. Identically,

y 0 = y 0 ∨ 0 = y 0 ∨ (x ∧ y) = (y 0 ∨ x) ∧ (y 0 ∨ y)
= 1 ∧ (y 0 ∨ y)
= y 0 ∨ y,

so that y ≤ y 0 . Therefore y = y 0 , so that the complement of x is indeed

unique.
Of course this means that in a complemented distributive lattice, each
element x has a unique complement, which we will denote by x∗ .
Definition 2.1.8. A Boolean algebra is a complemented distributive lattice.

2.2 Properties of a Boolean algebra

By the results and definitions in Section 2.1, the following hold for any x, y, z
in a Boolean algebra:

B1 x∨y =y∨x x∧y =y∧x

B2 x ∨ (y ∨ z) = (x ∨ y) ∨ z x ∧ (y ∧ z) = (x ∧ y) ∧ z
B3 (x ∧ y) ∨ y = y (x ∨ y) ∧ y = y
B4 x ∧ (y ∨ z) = (x ∧ y) ∨ (x ∧ z) x ∨ (y ∧ z) = (x ∨ y) ∧ (x ∨ z)
B5 x ∨ x∗ = 1 x ∧ x∗ = 0.

Theorem 2.2.1. Let hB, ∨, ∧, ∗ , 0, 1i be a structure consisting of a set B,

two binary operations ∨, ∧, one unary operation ∗ , and two designated el-
ements 0, 1 ∈ B. Suppose the identities B1-B5 hold in this structure and
define a relation ≤ on B by
x≤y iff x ∧ y = x.
Then hB, ≤i is a Boolean algebra [1].

10
Proof. First we must show that hB, ≤i is a lattice; to show that ≤ as defined
above is actually an order relation, first consider reflexivity:

x = x ∧ (x ∨ x∗ ) by B3
= (x ∧ x) ∨ (x ∧ x∗ ) by B4
= (x ∧ x) ∨ 0 by B5
= (x ∧ x) ∨ [(x ∧ x) ∧ (x ∧ x)∗ ] by B5
= [(x ∧ x) ∨ (x ∧ x)] ∧ [(x ∧ x) ∨ (x ∧ x)∗ ] by B4
   
= (x ∧ x) ∧ [(x ∧ x) ∨ (x ∧ x)∗ ] ∨ (x ∧ x) ∧ [(x ∧ x) ∨ (x ∧ x)∗ ] by B4

= (x ∧ x) ∨ (x ∧ x) by B3
= [(x ∧ x) ∨ x] ∧ [(x ∧ x) ∨ x] by B4
=x∧x by B3

Therefore we have x ≤ x. For antisymmetry, suppose that x ≤ y and y ≤ x.

Then we have two equations x ∧ y = x and y ∧ x = y. Of course from B1,
we have
x = x ∧ y = y ∧ x = y,
so ≤ is indeed antisymmetric. Finally for transitivity suppose that x ≤ y
and y ≤ z. Then x ∧ y = x and y ∧ z = y, so

x ∧ z = (x ∧ y) ∧ z = x ∧ (y ∧ z) = x ∧ y = x,

and we conclude x ≤ z. So hB, ≤i is a partially ordered set; since we are

guaranteed 0, 1 ∈ B we know that B is nonempty. Next we must show that
any two elements x, y ∈ B have both a supremum and infimum, namely
x ∨ y and x ∧ y respectively. To see that x ∧ y is a lower bound for {x, y},
recall first that in the proof of reflexivity, we showed x ∧ x = x. Hence by
B1 and B2,

(x ∧ y) ∧ x = x ∧ (x ∧ y) = (x ∧ x) ∧ y = x ∧ y,

so x ∧ y ≤ x. Obviously

(x ∧ y) ∧ y = x ∧ (y ∧ y) = x ∧ y

as well, so that x ∧ y ≤ y and therefore x ∧ y is a lower bound for {x, y}.

To see that it is the infimum, suppose z ∈ B is any lower bound for {x, y}.

11
This means that z ≤ x and z ≤ y, so we have the two equations z = x ∧ z
and z = y ∧ z. We wish to show that z ≤ x ∧ y, which follows easily:

z ∧ (x ∧ y) = (z ∧ x) ∧ y = z ∧ y = z,

or equivalently z ≤ x∧y. Therefore x∧y = inf{x, y}. The proof that x∨y =
sup{x, y} is very similar, so we omit it for the sake of brevity. We have now
shown that hB, ≤i is a lattice. The fact that this lattice is distributive follows
from B4. All that remains is to show that this lattice is complemented. This
requires least and greatest elements, which are predictably going to be the
designated elements 0 and 1. For any z ∈ B, recall that by reflexivity,
z = z ∧ z. Then

0 ∧ z = (z ∧ z ∗ ) ∧ z by B5
= z ∧ (z ∧ z ∗ ) by B1
= (z ∧ z) ∧ z ∗ by B2
= z ∧ z∗ by reflexivity
=0 by B5

so 0 ≤ z for all z ∈ B. Also

z ∧ 1 = z ∧ (z ∨ z ∗ ) by B5
=z by B3

shows that z ≤ 1 for all z ∈ B. The unary operator will yield complements:
for any z ∈ B we have the element z ∗ such that z ∨ z ∗ = 1 and z ∧ z ∗ = 0,
directly from B5. Therefore hB, ≤i is a complemented distributive lattice
which is of course a Boolean algebra.

The characterization in Theorem 2.2.1 will be used frequently, and will

be the key to viewing the propositional calculus as an algebraic structure.
We can also see from the symmetry in this characterization the principle
of duality, which is the observation that if a statement P is true about all
Boolean algebras, then the dual of P which is obtained from P by inter-
changing ∧ with ∨, and 0 with 1, is also true. We can see that the structure
described in Theorem 2.2.1 would obviously remain the same if the two
columns describing B1-B5 were swapped. This effectively allows us to cut
the number of our theorems in half!

12
Theorem 2.2.2. The principle of duality holds in all Boolean algebras.
Proof. This proof closely follows the argument in [1]. Suppose a statement
P holds in all Boolean algebras, and let hB, ≤i be any such Boolean algebra.
We will define a new relation ≤0 on B, such that
x ≤0 y iff y ≤ x.
This is evidently another partial ordering on B, where the join (meet) of x
and y with respect to ≤0 is precisely their meet (join) with respect to ≤. We
will show one of these facts to illustrate: to show that the new join x ∨0 y is
precisely the meet x ∧ y, first note that x ∨0 y = x ∧ y is in fact an upper
bound for {x, y} with respect to ≤0 . We have x ∧ y ≤ x and x ∧ y ≤ y, so
that x ≤0 x ∧ y = x ∨0 y and y ≤0 x ∧ y = x ∨0 y. Next let z be any upper
bound for {x, y} with respect to ≤0 . Then we have x ≤0 z and y ≤0 z, which
of course means that z ≤ x and z ≤ y. So z is a lower bound for {x, y} with
respect to ≤, so we must have that z ≤ x ∧ y. Then x ∨0 y = x ∧ y ≤0 z, so in
fact ∨0 = ∧ is well defined as a join operator with respect to ≤0 . A similar
argument shows that ∧0 = ∨ is well defined as a new meet operator with
respect to ≤0 . Also since 0 ≤ z ≤ 1 for all z ∈ B, we have 1 ≤0 z ≤0 0, so
that 1 and 0 are now the least and greatest elements with respect to ≤0 , so
we may as well call them 00 and 10 respectively. Complementation remains
the same; from B5 we have the least element 00 = 1 = x ∨ x∗ = x ∧0 x∗ , and
the greatest element 10 = 0 = x ∧ x∗ = x ∨0 x∗ .
Since P was assumed to hold in all Boolean algebras, it must hold in
hB, ≤0 i. But the coinciding of ∨0 = ∧, ∧0 = ∨, 00 = 1,and 10 = 0 means that
the dual of P must also hold in hB, ≤i.
To illustrate the convenience of duality, we will prove the following small
yet useful fact.
Proposition 2.2.3. For any elements x and y of a Boolean algebra B,
(x ∧ y)∗ = x∗ ∨ y ∗ .
Proof. We have

(x∗ ∨ y ∗ ) ∧ (x ∧ y) = [x∗ ∧ (x ∧ y)] ∨ [y ∗ ∧ (x ∧ y)]

= [(x∗ ∧ x) ∧ y] ∨ [(y ∗ ∧ y) ∧ x]
= [0 ∧ y] ∨ [0 ∧ x]
=0∨0
= 0.

13
Also

(x∗ ∨ y ∗ ) ∨ (x ∧ y) = [(x∗ ∨ y ∗ ) ∨ x] ∧ [(x∗ ∨ y ∗ ) ∨ y]

= [(x∗ ∨ x) ∨ y ∗ ] ∧ [x∗ ∨ (y ∗ ∨ y)]
= [1 ∨ y ∗ ] ∧ [x∗ ∨ 1]
=1∧1
= 1.

Since complements in B are unique, we have (x ∧ y)∗ = x∗ ∨ y ∗ .

Corollary 2.2.4. For any elements x and y of a Boolean algebra B,

(x ∨ y)∗ = x∗ ∧ y ∗ .

This follows immediately from Proposition 2.2.3 and the principle of du-
ality. For the remainder of this paper we will refer to duality very frequently;
for those who may be unsatisfied with the above proof of this principle, refer
to [2, Ch. 22] for a more detailed discussion.

2.3 Examples of Boolean Algebras

We have already seen that the power set P(X) of a set X under set inclusion
is a lattice; it is not too difficult to see that it is also a Boolean algebra.

Example 2.3.1. The power set lattice hP(X), ⊆i is a Boolean algebra,

where

A ∨ B = A ∪ B, A ∧ B = A ∩ B,
0 = ∅, 1 = X,

A∗ = X \ A.
Distribution of intersection over union and union over intersection are well
known set-theoretic facts. Complements of sets are also readily seen as
consistent with complementation as defined for a Boolean algebra.

Example 2.3.2. Define a total ordering ≤ on the two-element set 2 = {0, 1}

by
0 ≤ 0, 0 ≤ 1, 1 ≤ 1.

14
Then h2, ≤i is a Boolean algebra called the minimal algebra. This minimal
algebra will come up quite often in our talk of homomorphisms later on;
in fact, this “trivial” algebra is fundamental in our connection of Boolean
algebra to propositional calculus.
Finally, the most important example:
Example 2.3.3. Let L be the propositional language described in Section 1
and let S be the set of all sentences of L. Define the relation ≡ such that
for x, y ∈ S,
x≡y iff ` x ⇔ y.
Then ≡ is easily seen to be an equivalence relation: reflexivity follows from
the tautology x ⇒ x; symmetry follows from T2 by considering x ⇔ y =
(x ⇒ y) ∧ (y ⇒ x); transitivity follows directly from A4.
For each x ∈ S, let |x| denote the ≡-class of x, and let B = {|x| : x ∈ S}
be the set of all ≡-classes. Define the relation ≤ on B by
|x| ≤ |y| iff ` x ⇒ y,
and define join, meet, and complementation as follows:
|x| ∨ |y| = |x ∨ y|, |x| ∧ |y| = |x ∧ y|,
|x|∗ = |¬x|.
Assign the values 1 = |x| for any theorem x and 0 = |y| for any y such
that ¬y is a theorem. Note that above, although the symbols are identical,
the first instances of ∨ and ∧ are newly defined, acting as join and meet on
B, while the second occurrences are the propositional connectives defined
in Section 1. We will show that these operations and assignments are well
defined in the following theorem. As an aside, the fact that these operations
are well defined is similar to saying that ≡ is a congruence relation, and
thus B is a quotient algebra. The only reason this is not exactly the case
is because the set of all sentences isn’t originally endowed with the Boolean
operations and ordering, so to say elements in the equivalence classes are
congruent with respect to those operations is not a properly stated sentence;
however, the equivalence classes themselves now act as the elements of a
Boolean algebra, as stated in the following theorem.
Theorem 2.3.4. The set B with relations and operations defined above is
a Boolean algebra, called the Lindenbaum algebra 1 of L [1].
1
The Lindenbaum algebra, or Lindenbaum-Tarski algebra is named for logicians Adolf
Lindenbaum and Alfred Tarski. It was first introduced by Tarski in 1935 as a device to
establish correspondence between classical propositional calculus and Boolean algebras.
The Lindenbaum-Tarski algebra is considered the origin of the modern algebraic logic [5].

15
Proof. We hope the reader is satisfied with the informal argument of ≡
being an equivalence relation; to be sure, a rigorous proof would actually
rely on formal deductions as in Section 1, which as we’ve already mentioned
are usually more tedious than enlightening. With that fact in hand, the
next step in this proof is to show that ≤ is well defined. This means that if
x2 ∈ |x1 | and y2 ∈ |y1 |, then

|x1 | ≤ |y1 | implies |x2 | ≤ |y2 |,

or as we defined the relation above,

` x1 ⇒ y1 implies ` x2 ⇒ y2 .

But here, x1 ≡ x2 means precisely that ` x1 ⇔ x2 , and similarly ` y1 ⇔ y2 .

Again, technically a formal proof is necessary, but we will be informal for
the sake of brevity. From A4 we can obtain


` (x1 ⇒ y1 ) ⇒ (x2 ⇒ x1 ) ⇒ (x2 ⇒ y1 ) ,

from which a few applications of modus ponens yields ` x2 ⇒ y1 . Again an

instance of A4 is


` (y1 ⇒ y2 ) ⇒ (x2 ⇒ y1 ) ⇒ (x2 ⇒ y2 ) ,

and it follows from modus ponens that ` x2 ⇒ y2 . Therefore |x2 | ≤ |y2 | and
we conclude that ≤ is well defined.
Next we must show that the join operation is well defined. As above,
suppose that x1 ≡ x2 and y1 ≡ y2 . We must show that x1 ∨y1 ≡ x2 ∨y2 . Again
this is equivalent to showing that if x1 ⇔ x2 and y1 ⇔ y2 are theorems, then
so is x1 ∨ y1 ⇔ x2 ∨ y2 . As a few instances of A4 we have


` (y1 ⇒ y2 ) ⇒ (x2 ∨ y1 ) ⇒ (x2 ∨ y2 ) ,


` (x1 ⇒ x2 ) ⇒ (y1 ∨ x1 ) ⇒ (y1 ∨ x2 ) ,




` (x2 ∨y1 ) ⇒ (x2 ∨y2 ) ⇒ (x1 ∨y1 ) ⇒ (x2 ∨y1 ) ⇒ (x1 ∨y1 ) ⇒ (x2 ∨y2 ) .
Multiple applications of modus ponens and the commutativity in T1 will
yield that ` x1 ∨y1 ⇒ x2 ∨y2 . Switching around the indices in this argument
will also show that ` x2 ∨ y2 ⇒ x1 ∨ y1 , so that

` x1 ∨ y1 ⇔ x2 ∨ y2

which means x1 ∨ y1 ≡ x2 ∨ y2 , so ∨ is well defined.

16
 To show that complementation is well defined, suppose x ≡ y. Then
` x ⇔ y, and it follows from A3 that

x ⇒ y = ¬x ∨ y
⇒ y ∨ ¬x
= ¬y ⇒ ¬x,
y ⇒ x = ¬y ∨ x
⇒ x ∨ ¬y
= ¬x ⇒ ¬y,

and therefore from x ⇔ y it can be deduced that ¬x ⇔ ¬y. This means

that x ≡ y if and only if ¬x ≡ ¬y, so that |x|∗ = |¬x| is well defined.
To show that the meet operation is well defined, we rely on the fact that
∧ can be defined in terms of ∨ and ¬ in the propositional calculus. As above,
suppose that x1 ≡ x2 and y1 ≡ y2 . From above we know that ¬x1 ≡ ¬x2
and ¬y1 ≡ ¬y2 , and also

¬x1 ∨ ¬y1 ≡ ¬x2 ∨ ¬y2 .

By definition this means

x1 ⇒ ¬y1 ≡ x2 ⇒ ¬y2 .

Finally, again since complementation is well defined,

¬(x1 ⇒ ¬y1 ) ≡ ¬(x2 ⇒ ¬y2 ),

which by definition of ∧ over sentences means precisely that

x1 ∧ y1 ≡ x2 ∧ y2 .

We conclude that meet is also well defined.

Next we consider the definitions for the least and greatest element of B.
To show that these are well defined, we must prove that for any theorems x
and y, ` x ⇔ y. This shows that the greatest element is defined, and from
this and the fact that complementation is well defined, we can infer that
the least element is also well defined. First note that the axioms A2 and
A3 together imply that ` x ⇒ (y ⇒ x) for any sentence x. Thus for any
theorems x and y,
` x ⇒ (y ⇒ x),

17
 ` y ⇒ (x ⇒ y),
` x,
` y,
and a few applications of modus ponens will show that ` x ⇔ y, so that
|x| = |y|. As mentioned above, it follows that both the least and greatest
elements are well defined.
Now that we know the definitions given in Example 2.3.3 are valid, we
will use Theorem 2.2.1 to show that these operations form a Boolean algebra
on B. Note that the theorems proven in Section 1.4 coincide directly with
the properties B1-B4 from Theorem 2.2.1. For example the fact that T1 is
a theorem shows that x ∨ y ≡ y ∨ x, so that

|x ∨ y| = |y ∨ x|

and thus
|x| ∨ |y| = |y| ∨ |x|.
In this way, the properties B1-B4 are immediate. We know from Exam-
ple 1.2.2 that ` ¬x ∨ x, and from above we have

1 = |¬x ∨ x| = |x|∗ ∨ |x| = |x| ∨ |x|∗ .

Also note that since x ⇒ x is a theorem for any sentence x, replacing ¬x

for x shows that ¬x ⇒ ¬x is also a theorem. Hence

0 = |¬(¬x ⇒ ¬x)| = |¬x ∧ x| = |x|∗ ∧ |x| = |x| ∧ |x|∗ .

Therefore B5 is also satisfied, so B is a Boolean algebra.

Although the formalities in the syntax of propositional language may be

opaque, the coinciding of propositional connectives “or,” “and,” and “not”
with the algebraic operations join, meet, and complementation should be
intuitive. For instance, the familiar notion of double negation in a logic set-
ting would lead us to believe (or at least hope) that ` ¬¬x ⇔ x. Although a
formal proof could show this, the construction would be much more difficult
than such a small result deserves. However, this notion is rather obvious
when we think of sentences (to be precise, equivalence classes of provably
equivalent sentences) in the Lindenbaum algebra. Of course since comple-
mentation is unique in any Boolean algebra B, we must have x∗∗ = x for
any x ∈ B. In the Lindenbaum algebra this means that |x|∗∗ = |x|, which is
true if and only if ` ¬¬x ⇔ x. So, while the proof above may have seemed

18
so terribly cumbersome, this result really does allow us to use significant
machinery from algebra to simplify the concepts in a formal propositional
calculus.
Another familiar notion is the fact that anything is implied from a false
hypothesis. Formally, we might conjecture that ` (x ∧ ¬x) ⇒ y for any
sentences x and y. Of course, since 0 = |x| ∧ |x|∗ = |x ∧ ¬x| and 0 ≤ |y| for
all |y| ∈ B, we see that the conjecture is true. These facts not only satisfy
the craving of shorter and more elegant proofs, but also they help assure us
that we defined the propositional calculus correctly.
We are almost in a position to further explore the properties of the propo-
sitional calculus. The next section will again be in a general Boolean algebra
setting, and then we will return to the Lindenbaum algebra in Section 4.

3 Structure of a Boolean Algebra

For this section, let B be any Boolean algebra.

3.1 Homomorphisms
The following are familiar definitions in any study of algebra.
Definition 3.1.1. A Boolean homomorphism is a function h : B → B 0 for
Boolean algebras B and B 0 such that for all x, y ∈ B,

h(x ∨ y) = h(x) ∨ h(y),

h(x ∧ y) = h(x) ∧ h(y),

and
h(x∗ ) = h(x)∗ .
Be sure to recognize that we have used the universal Boolean symbols
for join, meet, and complementation, but really B and B 0 could be defined
with different operations. Thus in the definition above, the operations on x
and y are in B and those on h(x) and h(y) are in B 0 .
Definition 3.1.2. The kernel of a Boolean homomorphism f is the set

ker f = f −1 [{0}].

In words, the kernel is the subset of the domain that maps to the least
element in the codomain of f . We also define the hull of f as the set

hull f = f −1 [{1}].

19
Note that these two definitions are dual to one another. That is, if some-
thing is true about the set of elements mapping to the least element in the
codomain of f , then there will be a parallel result for the set of elements
mapping to the greatest element in the codomain of f .

Proposition 3.1.3. A homomorphism f : B → B 0 is one-to-one if and only

if either the kernel of f is trivial or the hull of f is trivial.

Proof. First note that we must have 0 ∈ ker f , because for any x ∈ B,

f (0) = f (x ∧ x∗ ) = f (x) ∧ f (x)∗ = 0.

Also we see as a result from the dualism between kernels and hulls that
x ∈ ker f if and only if x∗ ∈ hull f , because

f (x) = 0 iff f (x)∗ = 0∗ iff f (x∗ ) = 1.

Thus we see that since 0 is always in the kernel of a homomorphism, 1 is

always in the hull. Furthermore, this shows that the kernel of f is trivial if
and only if the hull of f is trivial.
(=⇒) Of course if f is one-to-one then there will only be one element in
the kernel, namely 0. Moreover 1 will be the only element in the hull.
(⇐=) Conversely, suppose ker f = {0}. As argued above, this implies
hull f = {1}. Then if f (a) = f (b), we have

0 = f (a) ∧ f (a)∗ = f (a) ∧ f (b)∗ = f (a ∧ b∗ ),

and it must be the case that a ∧ b∗ = 0. Similarly

1 = f (a) ∨ f (a)∗ = f (a) ∨ f (b)∗ = f (a ∨ b∗ ),

so we have a ∨ b∗ = 1 as well. Thus b∗ is the complement of a and since

complements are unique, we must have a = b so that f is one-to-one.

3.2 Ideals and Filters

The definitions in this section are mostly from [2]. A few useful and equiv-
alent definitions from [1] are noted as well.

Definition 3.2.1. A Boolean ideal is a subset I of B such that

(i) 0 ∈ I,

(ii) if x ∈ I and y ∈ I, then x ∨ y ∈ I,

20
(iii) if x ∈ I and y ∈ B, then x ∧ y ∈ I.

Definition 3.2.2. We define the dual concept of an ideal, called a Boolean

filter as a subset F of B such that

(i) 1 ∈ F ,

(ii) if x ∈ F and y ∈ F , then x ∧ y ∈ F ,

(iii) if x ∈ F and y ∈ B, then x ∨ y ∈ F .

A familiar algebraic result follows for Boolean algebra.

Proposition 3.2.3. The kernel of any Boolean homomorphism is a Boolean

ideal.

Proof. Let f : B → B 0 be a homomorphism for Boolean algebras B and B 0

and let I = ker f . As discussed in the proof of Proposition 3.1.3 we know
that 0 ∈ I. Next suppose that x, y ∈ I, which means that f (x) = f (y) = 0.
Then
f (x ∨ y) = f (x) ∨ f (y) = 0 ∨ 0 = 0,
so we have x ∨ y ∈ I as well. Finally if x ∈ I and z ∈ B, we have

f (x ∧ z) = f (x) ∧ f (z) = 0 ∧ f (z) = 0,

so x ∧ z ∈ I. Therefore ker f is an ideal.

Corollary 3.2.4. The hull of any Boolean homomorphism is a Boolean

filter.

The previous corollary follows directly from Proposition 3.2.3 and the
principle of duality. We note that if the reader is unsettled by using the
principle of duality blindly, one can easily typographically take the dual of
the proof above and see that it is a valid proof for the corollary. Of course in
taking the dual we replace the word “kernel” with “hull,” and “ideal” with
“filter.”
We pause to make a few observations of ideals and filters. If I is an
ideal and F is a filter, note that condition (iii) implies that whenever 1 ∈ I,
it must be the case that I = B, in which case we say I is improper. Also
whenever 0 ∈ F , we have F = B and F is improper.
Also a very important characterization (from the definition of ideals and
filters as given in [1]) is the fact that for ideals, condition (iii) can be replaced
by

21
 (iii) if x ∈ I and y ≤ x, then y ∈ I,

and similarly for filters,

(iii) if x ∈ F and x ≤ y, then y ∈ F .

This is because
y≤x iff y =y∧x
and
x≤y iff y = x ∨ y.
This characterization will be used frequently, as it begets a nice picture: in
the lattice B, ideals are built from the top down while filters are built from
the bottom up. One might say that ideals are closed downward while filters
are closed upward.
One important ingredient before discussing the implications of this deep
connection between Boolean algebra and propositional logic is familiarity
with the concept of a filter generated by a subset of B. Although the alge-
braist is probably more comfortable working with ideals, we will be working
with filters in the Lindenbaum algebra, and the reason will become clear as
we go on. We will mainly be interested in proper filters; note that from the
reasoning above, we can say that F is a proper filter in B if and only if

(i) 1 ∈ F ,

(ii) if x ∈ F and y ∈ F , then x ∧ y ∈ F ,

(iii) if x ∈ F and x ≤ y, then y ∈ F ,

(iv) 0 ∈
/ F.

With this characterization we are better equipped to discuss a filter

generated by a subset.

Definition 3.2.5. The filter F generated by a subset X of B is the smallest

filter containing X. If F is generated by a singleton, then F is said to be
principal.

This is a familiar definition, but the construction of this set in the next
theorem will be of great importance. The following definition and theorem
come from [1].

Definition 3.2.6. A subset X of B is said to have the finite meet property

if whenever x1 , x2 , . . . , xn ∈ X, we have x1 ∧ x2 ∧ · · · ∧ xn 6= 0.

22
 To make the following theorem less cluttered, we will adopt a notation
for finite joins and meets:
n
_
xi = x1 ∨ · · · ∨ xn
i=1

and
n
^
xi = x1 ∧ · · · ∧ xn .
i=1

Theorem 3.2.7. A subset X of a Boolean algebra B is included in some

proper filter in B if and only if X has the finite meet property.
Proof. The following argument closely follows that of [1].
(=⇒) Suppose X is included in a proper filter F . For any x1 , . . . , xn ∈ X,
we see that by condition (ii) we have

x1 ∧ x2 ∈ F,

(x1 ∧ x2 ) ∧ x3 ∈ F,
and inductively we see that x1 ∧ · · · ∧ xn ∈ F . Since F is proper it must be
the case that x1 ∧ · · · ∧ xn 6= 0, so X does have the finite meet property.
(⇐=) Conversely suppose that X has the finite meet property. If X is
empty then of course X is contained in every filter of B, so suppose X 6= ∅.
Define

X + = {y ∈ B : x1 ∧ · · · ∧ xn ≤ y for some x1 , . . . , xn ∈ X}.

Of course X ⊆ X + , and we will show that X + is indeed a proper filter,

namely the one generated by X. Recall the four properties above that
determine a proper filter. We have (i) 1 ∈ X + because we are consider-
ing X nonempty, so there must exist x ∈ X and of course x ≤ 1. To
show that (ii) X + is closed under meet, let y, y 0 ∈ X + . Then there exist
x1 , . . . , xn , x01 , . . . , x0m ∈ X such that
n
^ m
^
xi ≤ y and x0i ≤ y 0 .
i=1 i=1

Of course this means that

n
^  n
^
xi ∧ y = xi
i=1 i=1

23
and
m
^  m
^
x0i ∧ y 0 = x0i .
i=1 i=1

Therefore
n m n m
" # " #
^ ^    ^  ^ 
xi ∧ x0i ∧ y ∧ y0 = xi ∧ y ∧ x0i ∧ y 0
i=1 i=1 i=1 i=1
 n
^   m
^ 
= xi ∧ x0i
i=1 i=1
n
^ m
^
= xi ∧ x0i ,
i=1 i=1

which is true if and only if

n
^ m
^
xi ∧ x0i ≤ y ∧ y 0 .
i=1 i=1

Therefore y ∧ y 0 ∈ X + , so condition (ii) is satisfied. The fact that (iii) is

satisfied is immediate from the construction of X + . Finally since X has the
finite meet property, we know that x1 ∧ · · · ∧ xn 6= 0 for any x1 , . . . , xn ∈ X,
so that we always have 0 < x1 ∧ · · · ∧ xn . Hence 0 ∈ / X + and therefore X +
is a proper filter containing X, which completes the proof.

Of course X + is the smallest such filter containing X. If we consider

any filter F such that X ⊆ F , recall that filters are closed upward; to be
precise, F is closed under meet and contains all elements greater than any
given element in F . We see from the definition of X + that since X ⊆ F , it
must also be the case that X + ⊆ F . Thus we regard X + as a constructive
characterization for the filter generated by a nonempty set X. If X is empty
then of course the filter generated by X, which we recall is defined as the
smallest filter containing X, is simply the trivial filter {1}.

3.3 Ultrafilters
Those with a knowledge of abstract algebra recognize the importance of a
particular type of ideal called a maximal ideal. Just as filters are dual to
ideals, an ultrafilter is dual to a maximal ideal. For reasons we will see in the
following section, ultrafilters are yet another key component in connecting

24
the propositional calculus to Boolean algebra. The definitions and results
in this section are mainly from [1].

Definition 3.3.1. An ultrafilter is a proper filter F of a Boolean algebra B

that is not properly contained in any other proper filter. That is, if F 0 is a
filter and F ⊆ F 0 ⊂ B, then F = F 0 .

To guarantee the existence of such ultrafilters, we will show that any

proper filter F in B is contained in an ultrafilter. This will rely on Zorn’s
Lemma, which is stated below.

Zorn’s Lemma. If a nonempty partially ordered set (P, ≤) has the prop-
erty that every nonempty chain (totally ordered subset) has an upper bound
in P , then P contains at least one maximal element. That is, P contains an
element m such that there is no x ∈ P for which m < x.

Zorn’s Lemma is equivalent to the axiom of choice, and we will use it

freely.

Theorem 3.3.2. Each proper filter F in a Boolean algebra B is contained

in an ultrafilter.

Proof. Let F be the set of all proper filters containing F ; then F is partially
ordered under set inclusion, as in Example 2.1.2. Also since F contains
itself, F is nonempty. Just as in [1] and [4], we will invoke Zorn’s Lemma on
chains in F. We must show that chains have upper S bounds in this setting.
Let C be a nonempty chain in F, and let M = C. We will show that
M is a proper filter containing F and thus an upper bound for the chain C.
We already have 1 ∈ M because 1 is in every filter and we are considering
a nonempty chain. If x, y ∈ M then we must have x ∈ X and y ∈ Y for
some filters X and Y in C. Since C is a chain we can assume without loss
of generality that X ⊆ Y . Then x, y ∈ Y and since Y is a filter, we have
x ∧ y ∈ Y ⊆ M . Also if x ≤ z for some z ∈ B, we have z ∈ Y ⊆ M since
Y is a filter. To be sure that M is proper, note that since every filter A in
the chain is proper, we have 0 ∈ / A for all A ∈ C. Therefore 0 ∈ / M , which
shows that M is a proper filter and M is clearly an upper bound for C.
By Zorn’s Lemma, we conclude that F has at least one maximal element
U , which is a proper filter containing F that is not properly contained by any
other proper filter containing F . However we cannot immediately conclude
that U is an ultrafilter, because we have only shown that U is maximal
among those proper filters containing F , not among every possible filter in

25
B. However, supposing there is a proper filter U 0 in B such that U ⊆ U 0 ,
we see that U 0 must also contain F . Since U is maximal among the proper
filters containing F , we have U = U 0 . Therefore, by Definition 3.3.1, we
conclude that U is an ultrafilter containing F .

Corollary 3.3.3. Every nonzero element x ∈ B is included in some ultra-

filter U of B.

Proof. Note that for any nonzero x, the singleton {x} has the finite meet
property. Therefore by Theorem 3.2.7 we know there is a proper filter F
containing x and by Theorem 3.3.2 there is an ultrafilter U containing F ,
so that x ∈ U .

Proposition 3.3.4. If F is a filter in B and x ∈

/ F , then there exists an
ultrafilter U such that F ⊆ U and x ∈
/ U.

Proof. As above suppose F is a filter with x ∈

/ F . Then it must be the case
∗
that x ∧ y 6= 0 for all y ∈ F . To see why, suppose the contrary: if y ∈ F
and x∗ ∧ y = 0, then

x = x ∨ 0 = x ∨ (x∗ ∧ y)
= (x ∨ x∗ ) ∧ (x ∨ y)
= 1 ∧ (x ∨ y)
= x ∨ y,

which means that y ≤ x. But filters are closed upward, so this would imply
x ∈ F . So it must be the case that x∗ ∧ y 6= 0 for all y ∈ F . Thus we extend
F to a set
G = {z ∈ B : x∗ ∧ y ≤ z for some y ∈ F }.
Note that F ⊆ G because x∗ ∧ y ≤ y for any y ∈ F . Furthermore, we will
show that this set still has the finite meet property. Consider an arbitrary
collection z1 , z2 , . . . , zn ∈ G. For each zi we know that there exists yi ∈ F
such that x∗ ∧ yi ≤ zi . Then of course
n
^ n
^
∗
(x ∧ yi ) ≤ zi ,
i=1 i=1

26
and we find that
n
^ n
^  n
^ 
(x∗ ∧ yi ) = x∗ ∧ yi by B1 and B2
i=1 i=1 i=1
n
^
= x∗ ∧ yi
i=1
Vn
which is nonzero because i=1 yi ∈ F . Thus
n
^ n
^
∗
0 < x ∧ yi ≤ zi ,
i=1 i=1

which shows that G has the finite meet property. Thus by Theorem 3.2.7
there exists a proper filter H containing G and by Theorem 3.3.2 there exists
an ultrafilter U containing H, such that F ⊆ G ⊆ H ⊆ U for some ultrafilter
U . However we also have x∗ ∈ G from construction, so that x∗ ∈ U as well.
Then it must be the case that x ∈ / U , otherwise 0 = x ∧ x∗ ∈ U which would
contradict the ultrafilter being proper.

Corollary 3.3.5. Every proper filter F in B is precisely the intersection of

all the ultrafilters containing F .

Proof. The forward containment is trivial. For the backward containment,

let x be in the intersection of all ultrafilters containing F . Because F is
proper, this intersection is nonempty by Theorem 3.3.2. Well, if x is in every
ultrafilter containing F , then there does not exist an ultrafilter U such that
F ⊆ U and x ∈ / U . Hence by the contrapositive of Proposition 3.3.4, we
must have x ∈ F . Therefore the filter F is precisely the intersection of all
ultrafilters containing F .

Although these results are dense, we will use all of them in the following
section, and they will make many theorems about the propositional calculus
much more simple and elegant. This remark is similar to the one made at the
end of Section 2, but the machinery here will be used in proving theorems
about the propositional calculus, not the theorems within the propositional
calculus from Definition 1.2.1.2 The next lemma and theorem will look very
familiar to those who have studied maximal ideals in algebra, as the dualism
is transparent.
2
One might say that these results will be used in proving metatheorems about L, as
opposed to proving formal theorems within L.

27
Lemma 3.3.6. Let U be an ultrafilter in B. Then if x ∨ y ∈ U , it must be
the case that x ∈ U or y ∈ U .
Proof. This proof follows the reasoning in [1]. Let x, y ∈ B such that x ∨ y ∈
U . Suppose that x ∈/ U and construct the set
F = {z ∈ B : x ∨ z ∈ U }.
First note that since U is a filter, for any z ∈ U we have x ∨ z ∈ U , so
we know U ⊆ F . If we can show that F is a proper filter, we can conclude
U = F . There are four conditions to show. First we know that 1 ∈ F because
x ∨ 1 = 1 ∈ U . Next consider any two elements z1 , z2 ∈ F . Since x ∨ z1 and
x ∨ z2 are both in the the filter U , we must have (x ∨ z1 ) ∧ (x ∨ z2 ) ∈ U . But
by B4 we know that
x ∨ (z1 ∧ z2 ) = (x ∨ z1 ) ∧ (x ∨ z2 ),
and therefore z1 ∧ z2 ∈ F as well. Next let w ∈ B such that z ≤ w for some
z ∈ F . Then w = z ∨ w and x ∨ z ∈ U . But since U is a filter we know that
(x ∨ z) ∨ w ∈ U , and
(x ∨ z) ∨ w = x ∨ (z ∨ w) = x ∨ w.
Hence x ∨ w ∈ U so we conclude that w ∈ F . Finally, note that x ∨ 0 = x
and recall that x ∈/ U . Hence 0 ∈ / F and we conclude that F is a proper
filter. But F contains the ultrafilter U , so it must be the case that F = U .
Recall that x ∨ y ∈ U , from which it follows that y ∈ F . Of course since
F = U , we have that y ∈ U as well.

Theorem 3.3.7. F is an ultrafilter if and only if F is the hull of a homo-

morphism h : B → 2.
Proof. (=⇒) Suppose F is an ultrafilter. Then define h : B → {0, 1} by
(
0 if x ∈ /F
h(x) =
1 if x ∈ F.
Let x, y ∈ B. We will show that h(x ∨ y) = h(x) ∨ h(y) and h(x ∧ y) =
h(x) ∧ h(y) by considering three cases.
1. If both x, y ∈ F , then x ∨ y ∈ F and x ∧ y ∈ F so that
h(x ∨ y) = 1 = 1 ∨ 1 = h(x) ∨ h(y)
and
h(x ∧ y) = 1 = 1 ∧ 1 = h(x) ∧ h(y).

28
 2. If only one of x and y is in F , without loss of generality we will suppose
x ∈ F and y ∈ / F . Then x ∨ y ∈ F so that

h(x ∨ y) = 1 = 1 ∨ 0 = h(x) ∨ h(y).

However x ∧ y ∈
/ F , because x ∧ y ≤ y, and this would imply y ∈ F .
Hence
h(x ∧ y) = 0 = 1 ∧ 0 = h(x) ∧ h(y).

3. If both x, y ∈
/ F , then by the contrapositive of Lemma 3.3.6 we know
that x ∨ y ∈/ F . Therefore

h(x ∨ y) = 0 = 0 ∨ 0 = h(x) ∨ h(y).

Also we have x ∧ y ∈/ F , because if x ∧ y ∈ F , then x ∧ y ≤ y would

again imply y ∈ F . Thus

h(x ∧ y) = 0 = 0 ∧ 0 = h(x) ∧ h(y).

Finally we have that x ∈ F implies x∗ ∈ / F and x∗ ∈ F implies x ∈ / F ; either

implication immediately follows from considering the meet, as x ∧ x∗ = 0
can never be in an ultrafilter F . It follows that h(x∗ ) = h(x)∗ . Therefore we
conclude that h is a homomorphism and obviously by construction F is the
hull of h.
(⇐=) Suppose that F is the hull of a homomorphism h : B → 2, and let F 0
be a proper filter such that F ⊆ F 0 . Then if x ∈ F 0 , we must have x∗ ∈ / F,
otherwise x ∧ x∗ = 0 would be in F 0 , and F 0 would not be proper. Therefore
h(x∗ ) 6= 1, so that h(x)∗ = h(x∗ ) = 0 and hence h(x) = 1. Since F is the
hull of h we have x ∈ F so that F = F 0 . We conclude that F is indeed an
ultrafilter.

We now return at last to the Lindenbaum algebra B. The next section

will start by connecting the concept of a truth valuation to a homomorphism
on B, and adapting the definition of a deduction to apply to B. Afterward
we will see the simplification of some very important proofs regarding propo-
sitional calculus.

4 Applying Algebra
For this section, let B denote the Lindenbaum algebra as defined in Exam-
ple 2.3.3. If h : B → 2 is a homomorphism, we call it a 2-valued homomor-
phism.

29
4.1 Making Connections
Recall from Definition 1.3.1 that a truth valuation v is a function mapping
each sentence of L into {0, 1} such that for all sentences x and y,

v(¬x) = 1 if and only if v(x) = 0

v(x ⇒ y) = 1 if and only if v(x) = 0 or v(y) = 1.

It should be clear from this definition that the following properties are also
true:

v(x ∨ y) = 1 if and only if v(x) = 1 or v(y) = 1

v(x ∧ y) = 0 if and only if v(x) = 0 or v(y) = 0.

Before we move forward, we will need to show that truth valuations

in L correspond to homomorphisms in B, specifically that for any truth
valuation v, h(|x|) = v(x) is a well defined homomorphism onto the minimal
algebra, and also that a homomorphism onto the minimal algebra yields a
truth valuation. This will rely on the Weak Soundness Theorem, which we
will prove by induction. To do so, we need to formally define the length of
a deduction.
Definition 4.1.1. If x is a theorem, there exists a deduction of x. This is
a particular finite sequence of sentences

(y1 , y2 , . . . , yn ),

where yn = x. We define the length of a deduction as the number of

sentences in the deduction, so in this case the length of the deduction is n.
Note that there are actually many different deductions of a given theo-
rem; by the definition, any axioms can be added to the deduction (so long
as the finite sequence still ends with the given theorem) whether they are
necessary for the deduction or not. So there is not a single length assigned
for each theorem x, but a single length assigned to each deduction of x.
Theorem 4.1.2. If z is a sentence, then ` z implies  z.
Proof. Let (y1 , y2 , . . . , yn ) be a deduction of the theorem z. We proceed by
induction on the length n of this deduction. In the base cases when n = 1
or n = 2, the definition of a deduction implies that z is an axiom so it is
sufficient to check that each axiom is true in all truth valuations. So let v
be any truth valuation.

30
(A1) Suppose z = y ∨ y ⇒ y. If v(y ∨ y) = 0 then v(z) = 1. If instead
v(y ∨ y) = 1 we must have v(y) = 1 so that v(z) = v(y ∨ y ⇒ y) = 1.

(A2) Suppose z = x ⇒ x ∨ y. If v(x) = 0 then v(z) = 1. If instead v(x) = 1

we also have v(x ∨ y) = 1 so that v(z) = v(x ⇒ x ∨ y) = 1.

(A3) Suppose z = (x ∨ y) ⇒ (y ∨ x). Then if v(x ∨ y) = 0, we have v(z) = 1.

If instead v(x ∨ y) = 1 then we have v(x) = 1 or v(y) = 1, so that
v(y ∨ x) = 1. Hence

v(z) = v((x ∨ y) ⇒ (y ∨ x)) = 1

as well.

(A4) Suppose z = (x ⇒ y) ⇒ (w∨x ⇒ w∨y). If v(x ⇒ y) = 0 then we have

v(z) = 1. If instead v(x ⇒ y) = 1 then we have v(x) = 0 or v(y) = 1.
Consider w ∨ x; if v(w ∨ x) = 0 then we have v(w ∨ x ⇒ w ∨ y) = 1 so
that
v(z) = v((x ⇒ y) ⇒ (w ∨ x ⇒ w ∨ y)) = 1.
If instead v(w ∨ x) = 1 then we have v(w) = 1 or v(x) = 1. If v(w) = 1
then v(w ∨ y) = 1 and if v(x) = 1, from v(x ⇒ y) = 1 we must have
v(y) = 1 so that again v(w ∨ y) = 1. Thus v(w ∨ x ⇒ w ∨ y) = 1 so
that again we have

v((x ⇒ y) ⇒ (w ∨ x ⇒ w ∨ y)) = 1.

Thus all axioms are tautologies, and this covers the base case.
Next suppose that n > 2 and for all natural numbers k < n, if there
exists a deduction of a sentence y of length k, then  y. Now consider the
deduction of z which we denoted above as (y1 , . . . , yn ). Either z is an axiom,
which as we proved in the base case will imply that  z, or z follows from
modus ponens from two sentences earlier in the sequence, say

yj and yk = yj ⇒ z,

where j < n and k < n. However the sequences

(y1 , . . . , yj ) and (y1 , . . . yk )

are clearly deductions in their own right, with lengths j and k respectively.
Hence by the induction hypothesis, we know that  yj and  yk . Let v be
any truth valuation; we know from the previous sentence that v(yj ) = 1 and

31
v(yk ) = 1. But yk = yj ⇒ z, and since v(yj ) = 1, we must have v(z) = 1.
Since v was arbitrary we have  z.
Therefore by induction on the length of the deduction of theorem z, we
have shown that ` z implies  z.

Theorem 4.1.3. If h : B → 2 is a homomorphism, then v(x) = h(|x|) is

a truth valuation. Furthermore if v is a truth valuation, then h : B → 2
defined by h(|x|) = v(x) is a homomorphism.

The following proof necessarily conflates the “truth values” 0 and 1 with
the elements of the minimal algebra 2, which is not a problem but certainly
worth noting. Of course this doesn’t imply that the truth values from Section
1 inherently have these algebraic properties, but rather the definition of
a truth valuation restricts the function in such a way that the images of
sentences in {0, 1} behave like the minimal algebra.

Proof. (=⇒) Suppose that h is a homomorphism and as above, define

v(x) = h(|x|) as a function on the set of sentences of L. In this direction
it is clear that the function is well defined, because if a sentence x1 = x2 ,
obviously we have x1 ≡ x2 so that h(|x1 |) = h(|x2 |). So we just need to show
that v satisfies the truth valuation properties as in Definition 1.3.1. Since h
is a homomorphism, we have

v(¬x) = h(|¬x|) = h(|x|∗ ) = h(|x|)∗ = v(x)∗

in the minimal algebra, which of course means that v(¬x) = 1 if and only if
v(x) = 0. For the second property we also have

v(x ⇒ y) = h(|x ⇒ y|) = h(|¬x ∨ y|) = h(|x|)∗ ∨ h(|y|) = v(x)∗ ∨ v(y).

In this way we see that v(x ⇒ y) = 1 if and only if either v(x)∗ = 1 or

v(y) = 1, which is true if and only if v(x) = 0 or v(y) = 1. Therefore v is a
truth valuation.
(⇐=) Conversely, suppose that v is a truth valuation and let h(|x|) = v(x).
First let us show that h is well defined. So suppose that x1 ≡ x2 ; we wish to
show that h(|x1 |) = h(|x2 |), and to do so we mush show that v(x1 ) = v(x2 ).
Recall that x1 ≡ x2 means precisely that ` x1 ⇔ x2 . By Theorem 4.1.2 we
know that  x1 ⇔ x2 , so that v(x1 ⇔ x2 ) = 1. Then

1 = v(x1 ⇔ x2 )
= v((x1 ⇒ x2 ) ∧ (x2 ⇒ x1 )).

32
Suppose for contradiction that v(x1 ) 6= v(x2 ), and without loss of generality
we will suppose further that v(x1 ) = 1 and v(x2 ) = 0. Then v(x1 ⇒ x2 ) = 0,
so that

v (x1 ⇒ x2 ) ∧ (x2 ⇒ x1 ) = 0,
which is a contradiction. Thus we must have v(x1 ) = v(x2 ), and we conclude
that h(x) is well defined.
Next we must show that h is a homomorphism into 2. So let |x|, |y| ∈ B.
Then



h |x| ∨ |y| = h |x ∨ y| = v(x ∨ y)
(
1 if v(x) = 1 or v(y) = 1
= .
0 otherwise

We want the equation

h |x| ∨ |y| = h(|x|) ∨ h(|y|)

to hold true. We will consider the cases when v satisfies at least one of these
sentences, and when v does not satisfy either of them. In the first case,
suppose without
loss of generality that v(x) = 1. Then from above we see
that h |x| ∨ |y| = 1, and also in 2 we have

h(|x|) ∨ h(|y|) = v(x) ∨ v(y) = 1 ∨ v(y) = 1.

Otherwise
suppose both v(x) = 0 and v(y) = 0. Then from above we have

h |x| ∨ |y| = 0, and in 2 we have

h(|x|) ∨ h(|y|) = v(x) ∨ v(y) = 0 ∨ 0 = 0.

Thus in either case we have

h |x| ∨ |y| = h(|x|) ∨ h(|y|).

Next we have



h |x| ∧ |y| = h |x ∧ y| = v(x ∧ y)
(
0 if v(x) = 0 or v(y) = 0
= .
1 otherwise

Similar to above, we will consider the cases where v satisfies both x and
y, and where v does not satisfy one or both of them. In the first case, if

33


v(x) = v(y) = 1, then we have from above that h |x| ∧ |y| = 1. Also in 2
we have
h(|x|) ∧ h(|y|) = v(x) ∧ v(y) = 1 ∧ 1 = 1.
In the second case suppose without

loss of generality that v(x) = 0. Then
from above we have h |x| ∧ |y| = 0 and in 2 we have

h(|x|) ∧ h(|y|) = v(x) ∧ v(y) = 0 ∧ v(y) = 0.

Again in either case we find that

h |x| ∧ |y| = h(|x|) ∧ h(|y|).

Lastly we consider the preservation of negation. We have

(
0 if v(x) = 1
h(|x|∗ ) = h(|¬x|) = v(¬x) = .
1 if v(x) = 0

If v(x) = 1 then h(|x|∗ ) = 0 and also h(|x|)∗ = v(x)∗ = 1∗ = 0. Similarly if

v(x) = 0 then we have h(|x|∗ ) = 1 and also h(|x|)∗ = v(x)∗ = 0∗ = 1. Thus
we have
h(|x|∗ ) = h(|x|)∗ .
Therefore h(|x|) = v(x) is a well defined homomorphism for any truth val-
uation v. We have shown that each truth valuation gives rise to a 2-valued
homomorphism, and vice versa.

Now that we have the concept of truth valuations (as homomorphisms)

in B, we need to relate the process of deductions to B. Then we will have
a full arsenal of connecting semantic consequence to syntactic deductions in
our algebra, which will of course follow for propositional logic. We extend
Definition 1.2.3 to the Lindenbaum algebra with the following definition.

Definition 4.1.4. Let S be subset of B and |x| ∈ B. Then an algebraic

deduction of |x| from S is a finite sequence of elements of B such that each
element in the sequence is either (i) equal to 1, (ii) in S, or (iii) the result of
applying modus ponens to two earlier terms of the sequence. Here modus
ponens acts the same way: if |x| and |x ⇒ y| are two preceding terms in the
sequence, then |y| can be a subsequent term.

34
 In this setting we will be considering the set of all consequences Ŝ of S.
Note that since 1 is always in Ŝ, and

1 = |¬(x ∧ y) ∨ (x ∧ y)|
= |(x ∧ y)|∗ ∨ |x ∧ y|
= (|x| ∧ |y|)∗ ∨ |x ∧ y|
= (|x|∗ ∨ |y|∗ ) ∨ |x ∧ y|
= |x|∗ ∨ (|y|∗ ∨ |x ∧ y|)
= |x|∗ ∨ |y ⇒ (x ∧ y)|
 
= x ⇒ (y ⇒ (x ∧ y)),

we see that if |x| ∈ Ŝ and |y| ∈ Ŝ, then |x ∧ y| = |x| ∧ |y| ∈ Ŝ as well. Also
if for some |x| ∈ Ŝ we have |x| ≤ |y|, by definition this means that ` x ⇒ y,
which of course means that |x ⇒ y| = 1. Hence |x ⇒ y| ∈ Ŝ and therefore
|y| ∈ Ŝ by the modus ponens rule. These facts show us that Ŝ is a filter!
Explicitly we have shown

(i) 1 ∈ Ŝ,

(ii) if |x| ∈ Ŝ and |y| ∈ Ŝ then |x| ∧ |y| ∈ Ŝ,

(iii) if |x| ∈ Ŝ and |x| ≤ |y| then |y| ∈ Ŝ.

It is also clear that this is the smallest filter containing S, so that Ŝ is the
filter generated by S.
For the remainder of this paper, we will consider a set of sentences A in L
and define the corresponding subset SA = {|x| : x ∈ A} of the Lindenbaum
algebra B.
In the next few theorems we finally present the main results of this
paper, but first we need to assure ourselves that the existence of an algebraic
deduction in B of |y| from SA as in Definition 4.1.4 will imply the existence of
a deduction in L of y from A, as in Definition 1.2.3. Consider some algebraic
deduction (|x1 |, |x2 |, . . . , |xn |) of |y|, and let us analyze Defintion 4.1.4. For
any of the |xi | that are equal to 1, we know that by definition, ` xi . Thus
a formal deduction in L of y could include the deduction of xi of some
length li . If |xj | ∈ SA then we have xk ∈ A where xk ≡ xj . By definition,
` xk ⇔ xj , so we could include a formal deduction in L of xk ⇔ xj , and thus
with xk ∈ A, a formal deduction of xj from A of some length lj exists and

35
could be included in the deduction of y. Furthermore, if any |xi | followed
by modus ponens from two preceding terms |xj | and |xk | = |xj ⇒ xi |, by
the argument just given we could inductively deduce both xj and xj ⇒ xi
from A, resulting in a deduction of xi from A. Therefore, for any formula
|y| deduced algebraically from SA , there exists a deduction in L (albeit
likely much longer) of y from A. It is easy to see the converse is also
true, that if there is a formal deduction in L of y from A, then there is
an algebraic deduction (perhaps a quicker one) in B of |y| from SA . In fact
in this direction, we could just modify the terms of a deduction (y1 , . . . , yn )
simply by taking the equivalence classes (|y1 |, . . . , |yn |), which is evidently
an algebraic deduction.
With this assertion in hand, we begin to see the connections. A num-
ber of familiar concepts in propositional calculus can now be viewed alge-
braically. From the reasoning above, letting A be a set of sentences of L
and again SA = {|x| : x ∈ A}, we see that A ` y if and only if |y| is in the
filter generated by SA , that is |y| ∈ ŜA . This is the crux of the connection
between Boolean algebra and propositional logic – the algebraic properties
that bring about the closure of a filter in B correspond exactly to the rules
by which a theorem can be deduced from a set of sentences in L.

4.2 Soundness, Completeness, and Compactness

Readers with a previous knowledge of propositional logic will find the fol-
lowing results familiar, but the proofs are certainly enlightening.

Theorem 4.2.1. A set of sentences A of L is satisfiable if and only if it is

consistent.

Proof. (=⇒) Suppose A is satisfiable, which means there exists a truth

valuation v such that v[A] = {1}. Then the corresponding homomorphism h :
B → 2 defined by h(|x|) = v(x) (which is guaranteed by Theorem 4.1.3) will
map the equivalence classes of the elements of A to 1 as well. Equivalently,
h[SA ] = {1}. Let U = hull h, so that SA ⊆ U . Then U is an ultrafilter
by Theorem 3.3.7, so the smallest filter ŜA containing SA must also be
contained by U . Then since 0 ∈ / U , it must be the case that 0 ∈/ ŜA . Thus
∗
for any sentence x of L, since |x∧¬x| = |x|∧|x| = 0, we have |x∧¬x| ∈ / ŜA .
From above, this means that A 0 x ∧ ¬x, so that A is consistent.
(⇐=) The converse is similarly proved. We know A is consistent if and
only if for any sentence x, A 0 x ∧ ¬x. By the assertion above, we know
that A 0 x ∧ ¬x if and only if |x ∧ ¬x| ∈ / ŜA . But of course, |x ∧ ¬x| =
∗
|x| ∧ |x| = 0. Hence this is equivalent to saying that ŜA is a proper filter,

36
and by Theorem 3.3.2 we know there exists an ultrafilter U such that SA ⊆
ŜA ⊆ U . Then by Theorem 3.3.7 we know that U is the hull of some 2-
valued homomorphism h, so that h[SA ] = {1}. Then the truth valuation
v(x) = h(|x|) guaranteed by Theorem 4.1.3 is a truth valuation satisfying
A.

The theorem above is phrased in a propositional logic setting, but the

proof actually covers another equivalent condition for a set of sentences A
to be consistent and satisfiable. Note that in both directions we came across
the fact that 0 ∈/ ŜA , which means that ŜA is a proper filter. As we saw in
Theorem 3.2.7, this is equivalent to SA having the finite meet property. So
we have also shown that a set of sentences A in L is consistent (and thus
satisfiable) if and only if the corresponding subset SA has the finite meet
property in B.

Corollary 4.2.2. Let A be a set of sentences in L. The following are

equivalent:

(i) A is consistent

(ii) A is satisfiable

(iii) SA = {|x| : x ∈ A} has the finite meet property in B.

Proof. We need only show (i)⇔(iii). Suppose SA has the finite meet property
in B. By Theorem 3.2.7 this is true if and only if ŜA is a proper filter. As
stated in Section 3.2, ŜA is proper if and only if 0 ∈
/ ŜA . For any sentence
x, we have 0 = |x| ∧ |x|∗ = |x ∧ ¬x|, so this means that A 0 x ∧ ¬x for any
sentence x. By definition this means that A is consistent.

A remark was made in Section 3.2 that we are mainly interested in

proper filters, and this is precisely the reason; an inconsistent set of sentences
implies anything and is mathematically dull.
We can now prove a variation of the Deduction Theorem3 for proposi-
tional calculus.

Theorem 4.2.3. A ` y if and only if A has a finite subset {x1 , . . . , xn } such

that ` (x1 ∧ · · · ∧ xn ) ⇒ y.
3
This is actually a unique version of this theorem that befits this paper - it is modified
from [2].

37
Proof. Again let SA = {|x| : x ∈ A} and note that A ` y if and only if
|y| ∈ ŜA . As we saw in the constructive characterization in Theorem 3.2.7,
|y| ∈ ŜA if and only if there exist |x1 |, . . . , |xn | ∈ SA such that

|x1 | ∧ · · · ∧ |xn | ≤ |y|.

Since
|x1 | ∧ · · · ∧ |xn | = |x1 ∧ · · · ∧ xn |,
by definition this means that

` (x1 ∧ · · · ∧ xn ) ⇒ y.

The next theorem combines the Strong Soundness and Strong Complete-
ness Theorems.

Theorem 4.2.4. For any sentence y and set of sentences A in L, A ` y if

and only if A  y.

Proof. Again let SA = {|x| : x ∈ A}.

(=⇒) (Soundness) Suppose A ` y, which means |y| ∈ ŜA . If A is not
satisfiable, then by Definition 1.3.5, A  y is vacuously true. If A is satis-
fiable, then let v be any truth valuation satisfying A, so that v[A] = {1}.
Then by Theorem 4.1.3 there is a corresponding 2-valued homomorphism
h(|x|) = v(x) so that h[SA ] = {1}. This means that SA ⊆ hull h, and since
this hull is a filter and ŜA is the smallest filter containing SA , we must have
ŜA ⊆ hull h as well. Thus since |y| ∈ ŜA , it must be the case that h(|y|) = 1.
Then of course v(y) = 1 as well, and since v was arbitrary we conclude
A  y.
(⇐=) (Completeness) Suppose A  y. If ŜA is improper, then of course
|y| ∈ ŜA so that A ` y. Otherwise if ŜA is proper, let U be any ultrafilter
containing ŜA . At least one such ultrafilter exists by Theorem 3.3.2. By
Theorem 3.3.7 we know U is the hull of some 2-valued homomorphism h.
Of course since SA ⊆ ŜA we know that SA ⊆ U as well, so that h[SA ] = {1}.
By Theorem 4.1.3 there is a corresponding truth valuation v(x) = h(|x|)
such that v[A] = {1} and since A  y, we have v(y) = 1. Then it must
be the case that h(|y|) = 1 as well, so that |y| ∈ U . Since U was an
arbitrary ultrafilter containing ŜA , we have that |y| is in the intersection of
all ultrafilters containing ŜA . Therefore by Corollary 3.3.5 we conclude that
y ∈ ŜA . Hence A ` y.

38
 We can also prove the Compactness Theorem through such algebraic
means. There are two equivalent statements of Compactness, and there are
enlightening approaches to each proof.
Theorem 4.2.5. For any sentence y and set of sentences A in L, we have
A  y if and only if there is a finite subset {x1 , . . . , xn } of A such that
{x1 , . . . , xn }  y.
Proof. (=⇒) Suppose that A  y. By the Completeness Theorem we have
that A ` y, which means that |y| ∈ ŜA . As seen in Theorem 3.2.7 there is a
finite subset {|x1 |, . . . , |xn |} of SA such that
|x1 | ∧ · · · ∧ |xn | ≤ |y|.
Of course this implies that |y| is in the filter generated by {|x1 |, . . . , |xn |},
so that {x1 , . . . , xn } ` y. By the Soundness Theorem, it follows that
{x1 , . . . , xn }  y.
(⇐=) Conversely, suppose that there is a finite subset {x1 , . . . , xn } of A
such that {x1 , . . . , xn }  y. From the Completeness Theorem we know
{x1 , . . . , xn } ` y which means |y| is in the filter F generated by {|x1 |, . . . , |xn |}.
Since {|x1 |, . . . , |xn |} ⊆ SA , it is clear that F ⊆ ŜA so that |y| ∈ ŜA . Hence
A ` y and again from Soundness we conclude A  y.

In the context of propositional logic, the previous theorem is easily

proven to be equivalent to the following theorem. However again using
algebraic techniques, the proof is even simpler.
Theorem 4.2.6. A set of sentences A in L is satisfiable if and only if every
finite subset of A is satisfiable.
Proof. The forward direction is quite obvious; if a truth valuation v satisfies
A, then v will also satisfy every finite subset of A. On the other hand,
suppose that every finite subset of A is satisfiable. As we stated at the
end of Theorem 4.2.1, this means that every finite subset of SA has the
finite meet property. Then obviously SA has the finite meet property as
well, because for any finite number of elements x1 , . . . , xn ∈ SA , the subset
{x1 , . . . , xn } has the finite meet property so that x1 ∧ · · · ∧ xn 6= 0. Therefore
we know that A is consistent and satisfiable.

5 Conclusion
We have found that many common notions in propositional calculus are just
disguised, yet familiar algebraic notions. Although it may seem that the

39
necessary machinery in Boolean algebra is too much trouble for such results
that could be proven strictly within propositional calculus, the insight to
these connections is certainly enlightening. Furthermore, theorems such as
Completeness and Compactness are significantly more difficult in a first-
order logic, but the theory of ultrafilters in a Boolean algebra extends quite
easily into the first-order logic setting. Thus the main ideas presented in
this paper actually have much deeper implications than the few proofs in
Section 4.2, and establish a great introduction to further work in Boolean
algebra applied to first-order logic.

40
References
[1] Bell, J. L., and Moshe Machover. A Course in Mathematical Logic.
Amsterdam: North-Holland Pub. Co., 1977.

[2] Halmos, Paul R., and Steven R. Givant. Logic as Algebra. 21 Vol. Wash-
ington, D.C.: Mathematical Association of America, 1998.

[3] Leary, Christopher. A Friendly Introduction to Mathematical Logic. 1

ed. Upper Saddle River: Prentice Hall, 1999.

[4] Fraleigh, John B. A First Course in Abstract Algebra. 7 ed. Reading,

Mass: Addison-Wesley Pub. Co., 1976.

[5] “Lindenbaum-Tarski Algebra.” Wikipedia, The Free Encyclopedia.

Wikimedia Foundation, Inc. 11 May 2012.

41