Chapter 1

Propositional Logic

A logic is defined by its syntax and semantics. The syntax of a logic defines the
class of statements that can be used to convey information. The semantics of a
logic defines properties of logical statements, such as satisfiability and validity,
in addition to relationships among these statements, such as equivalence and
implication. The goal of this chapter is to define the syntax and semantics of
propositional logic: one of the simplest and most commonly used logics.

1.1 Syntax
Propositional logic statements are formed using a set of propositional variables,
P1 , . . . , Pn . These variableswhich are also called Boolean variables, proposi-
tional symbols, atomic propositions, and atomsare variables that assume one
of two values, typically indicated by true and false. The simplest statement
one can write in propositional logic has the form Pi . It is called an atomic
statement and is interpreted as saying that variable Pi takes on the value true.
More complex statements in propositional logic are formed using logical con-
nectives. Specifically, sentences in propositional logic are formed according to
the following rules:
Every propositional variable is a sentence.
If and are sentences, then , , and are also sentences.
The symbols , and are called logical connectives and they stand for nega-
tion, conjunction and disjunction, respectively. Other connectives can also be
introduced such as implication and equivalence but these are defined in
terms of the three primitive connectives given above. In particular, the sentence
is a shorthand for . Similarly, the sentence is a shorthand for
() ().
An important syntactic notion in propositional logic is the propositional lit-
eral, which is a propositional variable Pi , called a positive literal, or the negation
of a propositional variable Pi , called a negative literal.

1
2 Class Notes for CS264A, UCLA

A B

X Y

Figure 1.1: A digital circuit.

Propositional knowledge bases

Another important syntactic notion is the propositional knowledge base, which
is a set of propositional sentences = 1 , 2 , . . .. The knowledge base is
a shorthand for the sentence 1 2 . . ., which is formed by conjoining all
sentences in the knowledge base. Notating such a potentially large sentence
using a set of smaller sentences is quite convenient as it allows us to break up
the large sentence into pieces that are written on multiple lines:

1
2
..
.

Consider now the digital circuit in Figure 1.1, which has two inputs and one
output. Suppose that we want to write down a propositional knowledge base
that captures our knowledge, or some of it, about the behavior of this circuit.
The very first step we have to consider is that of choosing the set of propositional
variables. There are a number of choices here. First, we may take the set of
propositional variables to be 1 = A, B and write the simple knowledge base:

1 = A B

to describe the circuit behavior. In this case, we are asserting conditions on

inputs A and B under which the circuit would output a high value.
Another knowledge base one can write involves the set of propositional vari-
ables 2 = A, B, C:

A B C
2 =
(A B) C,

which includes a propositional variable for the circuit output.

 c 2004
Adnan Darwiche 3

A third knowledge base one can write involves the set of propositional vari-
ables 3 = A, B, C, X, Y :


A X



A X

AB Y
3 =

(A B) Y



X Y C

(X Y ) C,
which includes propositional variables for the internal wires in the circuit. Ob-
viously, these knowledge base vary in the amount of information they convey,
where knowledge base 3 is the most informative, yet also the most complex.
Which one of these knowledge bases is most suitable depends on the application
we have in mind. For example, if our goal is to check whether two circuits over
inputs A and B have the same functionality, then knowledge base 1 is sufficient
since it completely characterizes the functionality of given circuit. However, if
we are interested in reasoning about the internal behavior of the circuitsuch as
whether X being high implies that Y is lowthen knowledge base 3 is what
we need since the other two knowledge bases do not even allow us to phrase
logical questions about the state of internal wires in the circuit.
The main point behind the previous example is that the choice of proposi-
tional variables is quite important, and must be made in the context of given
application, which in turns fixes the kind of logical questions one may want to
pose with respect to the knowledge base.

1.2 Semantics
The syntax of propositional logic specifies rules for composing grammatically
correct sentences. The semantics of propositional logic defines logical properties
of such sentences, including consistency and validity, and logical relationships
among sentences, including implication and equivalence.
These logical properties and relationships are easy to figure out for simple
sentences. For example, most people would agree that
A A is inconsistent,
A A is valid,
A and AB imply B, and
A B is equivalent to B A.
Yet, it may not be as obvious that AB and BA are equivalent, or that
(AB)(AB) implies A.1 For this reason, one needs a formal definition of
logical properties and relationship, especially if one is to automate the decision
of these notions.
1 It is not uncommon for people to claim that (AB) (AB) is inconsistent, which is

not correct.
4 Class Notes for CS264A, UCLA

1.2.1 Truth at a world

The semantics of propositional logic hinges on a key relationship between sen-
tences and worlds. A world is a function that assigns a value of true/false
to each propositional variable Pi , where (Pi ) denotes the value assigned by
world to variable Pi . Intuitively, a world represents a particular state of af-
fairs in which the value of each propositional variable is known. The semantics
of propositional logic is based on a simple definition which tells us whether a
sentence is true at a particular world , which is written as w |= . This
definition is given inductively as follows:

|= Pi iff (Pi ) = true.

|= iff 6|= .

|= iff |= or |= .

|= iff |= and |= .

A world is also called a truth assignment or an interpretation. Moreover, the

relationship |= has many alternative readings: satisfies ; entails ;
is satisfied by ; holds at ; in addition to is true at . We should note
here that deciding the truth of a sentence at a particular world is quite easy
algorithmically as it can be done in time which is linear in the size of sentence.

1.2.2 Logical properties

We are now ready to define the most central logical property of sentences, that of
consistency. Specifically, we say that sentence is consistent iff there is at least
one world at which is true, otherwise it is inconsistent. It is also common to
use the terms satisfiable/unsatisfiable instead of consistent/inconsistent, respec-
tively.2 The property of satisfiability is quite important since many other logical
notions can be reduced to satisfiability. Moreover, deciding the satisfiability of
a propositional sentence is the first problem proven to be NPcomplete.
Consider now a situation involving earthquakes, burglaries and alarms, lead-
ing to three propositional variables:

Earthquake: An earthquake took place.

Burglary: A burglary took place.

Alarm: The alarm was triggered.

This set of propositional variables leads to eight possible worlds:

2 Hence, satisfiability is both a property and a relationship. As a property, satisfiability ap-

plies to sentences. And as a relationship, satisfiability relates a world to a sentence. Moreover,

we have defined the satisfiability property in terms of the satisfiability relationship.
 c 2004
Adnan Darwiche 5

world Earthquake Burglary Alarm

1 true true true
2 true true false
3 true false true
4 true false false
5 false true true
6 false true false
7 false false true
8 false false false

In general, a set of n propositional variables leads to a total of 2n possible

worlds. Given the above set of worlds, we have the following:
Earthquake is true at 1 , . . . , 4 .
Earthquake is true at 5 , . . . , 8 (the complement of worlds 1 , . . . , 4 ).
Burglary is true at 3 , 4 , 7 , 8 .
Alarm is true at 1 , 3 , 5 , 7 .
(Earthquake Burglary) is true at 7 , 8 .
(Earthquake Burglary) Alarm is true at 1 , 3 , 5 , 7 , 8 .
(Earthquake Burglary)Alarm is true at 1 , 3 , 5 , 7 , 8 .
Burglary Burglary is not true at any world, hence, it is not satisfiable.
The symbol false is often used to denote a sentence which is unsatisfiable. Note
that we have also used false to denote one of the values that propositional
variables can assume. The symbol false is therefore overloaded in propositional
logic.
We now turn to another logical property, that of validity. Specifically, we say
that sentence is valid iff it is true at every world. For example, AA is valid,
and so is (A (AB))B. It is not uncommon for one to confuse invalidity
and inconsistency even though these are very distinct notions. If the sentence
is invalid, then one can identify a world at which is false. That does not
imply though that is false at every other world, in which case would be
inconsistent. Hence, inconsistency is a stronger property than invalidity. Note,
however, that a sentence is valid iff its negation is inconsistent. This follows
because for every world , either is true at , or is true at , but not
both. Hence, the worlds at which is true and those at which is true form
a partition of the whole set of worlds.
Going back to the above example, the following sentences are valid:
Earthquake Earthquake.

Alarm((Earthquake Burglary)Alarm).
6 Class Notes for CS264A, UCLA

The symbol true is often used to denote a sentence which is valid.3 Moreover,
it is common to write |= in order to indicate that sentence is valid.
We close this section by defining a third logical property of sentences, that
of completeness. Specifically, we say that a sentence is complete iff it is true
at exactly one world. Intuitively, a complete sentence is very affirmative as
it communicates knowledge of the exact state of the world. Going back to our
previous example, the sentence EarthquakeBurglary is not complete as it is true
at six possible worlds, 1 , . . . , 6 . The sentence Earthquake Burglary Alarm
is complete though as it is true at exactly one world, 7 .

1.2.3 Logical relationships

A logical property applies to a single sentence, while a logical relationship applies
to two or more sentences. We now define a few logical relationships among
propositional sentences:
Two sentences are equivalent iff they are true at the same set of worlds.
Two sentences are mutually exclusive iff they are never true at the same
world.
Two sentences are exhaustive iff one of them is true at each world.
A fourth logical relationship which deserves special attention is that of implica-
tion. Intuitively, a sentence implies another sentence if in every situation
where is true, must also be true. In propositional logic, situations are
represented by worlds. Hence,
Sentence implies sentence iff whenever is true at world , then is
also true at world . That is, if |= then |= for all worlds .
We have earlier used the symbol |= to denote the satisfiability relationship
between a world and a sentence. Specifically, we wrote |= to indicate that
world satisfies sentence . This symbol is also used to indicate implication
between sentences, where we write |= to say that sentence implies sentence
. As an example statement which mixes the two uses of this symbol, we provide
again the definition of implication:
|= iff (if |= , then |= for every world ).
We observe again that either a sentence or its negation must be true at
each world : |= or |= . This property does not hold for implication,
however, as it is perfectly possible that we have 6|= and 6|= , that is,
neither implies nor implies . In fact, this property holds for implication
only when the sentence is complete (that is, is true at exactly one world).
In this case, we must have |= or |= for every sentence . We also
note that the implication relation among sentences is also known as entailment.
That is, we say that entails to mean that implies .
3 Again, we are overloading the symbol true since it also denotes one of the values that a

propositional variable may assume.

 c 2004
Adnan Darwiche 7

1.3 Knowledge as a set of possible worlds

The set of worlds that satisfy a sentence are called the models of and denoted
by Mods() = { : |= }.
Using the definition of satisfaction (|=), it is not hard to prove that the
function Mods satisfies the following properties:

Mods( ) = Mods() Mods().

Mods( ) = Mods() Mods().

Mods() = Mods().

The function Mods is very useful. First, it allows us to define logical notions
tersely. If denotes the set of all worlds, then

Sentence is satisfiable iff Mods() 6= .

Sentence is valid iff Mods() = .

Sentence is complete iff Mods() = {}.

Sentence implies sentence iff Mods() Mods().

Sentence is equivalent to sentence iff Mods() = Mods().

Two sentences and are mutually exclusive iff Mods()Mods() = .4

Two sentences and are exhaustive iff Mods() Mods() = .5

Second, the function Mods plays an important role in providing insights into
the semantics of propositional logic. The key concept here is to view ones
knowledge as a set of worlds that characterize all possibilities. In a complete
state of ignorance, every world is a possibility and, hence, our knowledge consists
in the set of all worlds . As we know more, some of these worlds are deemed as
impossible and, hence, the set of possible worlds starts shrinking. Specifically,
when someone communicates sentence to us, they are telling us that only
worlds in Mods() are possible, and every world outside Mods() is impossible.
Therefore, our state of knowledge in this case is Mods(). If we learn in
addition, our state of knowledge will be Mods( ) = Mods() Mods()
since only these worlds are possible, and every other world is impossible. We
reach a complete state of knowledge when we deem all worlds as impossible,
except for exactly one world.
Let us illustrate this concept of knowledge as a set of possible worlds using
the earthquakeburglaryalarm example that we introduced earlier. There are
eight worlds in this example, corresponding to eight possible ways the world
4 This can be generalized to an arbitrary number of sentences as follows. Sentences
1 , . . . , n are mutually exclusive iff Mods(i ) Mods(j ) = for i 6= j.
5 This can be generalized to an arbitrary number of sentences as follows. Sentences
1 , . . . , n are exhaustive iff Mods(1 ) . . . Mods(n ) = .
8 Class Notes for CS264A, UCLA

could be. For example, the second world 2 corresponds to the situation where
we have an earthquake and a burglary, but the alarm did not go off. In a
complete state of ignorance about this example, all of these eight situations are
possible. Hence, our state of knowledge is the set of all eight worlds 1 , . . . , 8 .
Suppose now that someone communicates to us the following sentence,

: (Earthquake Burglary)Alarm.

By accepting , we are considering some of these eight worlds as impossible. In

particular, any world that does not satisfy is ruled out. Therefore, our state
of knowledge now corresponds to the set,

Mods() = {1 , 3 , 5 , 7 , 8 }.

This is depicted in the following table, in which the sentence can be viewed
as having ruled out all worlds that are outside Mods():

world Earthquake Burglary Alarm Possible?

1 true true true yes
2 true true false
3 true false true yes
4 true false false
5 false true true yes
6 false true false
7 false false true yes
8 false false false yes

Suppose now that we also learn,

: EarthquakeBurglary,

for which Mods() = {1 , 2 , 5 , 6 , 7 , 8 }. Our state of knowledge is now

given by,

Mods( ) = Mods() Mods() = {1 , 5 , 7 , 8 },

which is one less world than it used to be. This is depicted by the following
table:

world Earthquake Burglary Alarm Possible?

1 true true true yes
2 true true false
3 true false true
4 true false false
5 false true true yes
6 false true false
7 false false true yes
8 false false false yes
 c 2004
Adnan Darwiche 9

Schema Equivalent Schema Name

false false
true
false
true true
double negation
( ) de Morgan
( ) de Morgan
( ) ( ) ( )
contraposition
( ) ( )
definition of
() () definition of

Table 1.1: Some equivalences among sentence schemas. The last two can be
viewed as definitions of the and connectives.

According to the view of knowledge as a set of possible worlds, the more

we know, the smaller the set of possible worlds becomes. Let us now recall
the definition of implication. A knowledge base implies a sentence iff
Mods() Mods(). Hence, the more information we have in , the smaller
the set Mods() is, and the more sentences are implied by . The extreme
case is when the knowledge base is unsatisfiable, Mods() = , leading to
|= for every sentence . Inconsistent knowledge bases are then useless since
they imply everything, including a sentence and its negation. Note also that if
is implied by , then adding to does not change our state of knowledge
since Mods( ) = Mods() in this case. Therefore, adding an implied sen-
tence to a knowledge base does not change the informational content of that
knowledge base.

1.4 Some equivalences among sentences

We now consider some equivalences between propositional sentences, which can
be quite useful when working with propositional logic. The equivalences are
given in Table 1.1 and are actually between schemas, which are templates that
can generate a large number of specific sentences (instances). For example,
is a schema, generating instances such as A(B C), where is replaced
by A and is replaced by (B C).
It is worth noting here that the first two equivalences on Table 1.1 follow as
a special case of the following more general rule:
If |= , then is equivalent to .
Similarly, the third and fourth equivalences in Table 1.1 follow from the following
rule:
10 Class Notes for CS264A, UCLA

Mods ( )

Mods( )

Figure 1.2: The relationship between the models of and assuming that
implies .

Relationship Property
implies is unsatisfiable
implies is valid
and are equivalent is valid
and are mutually exclusive is unsatisfiable
and are exhaustive is valid

Table 1.2: Some reductions among logical relationships and properties.

If |= , then is equivalent to .

These equivalences and others can be usually proven using the Mods function
and its properties. We will now provide an example proof of the following: If
|= , then is equivalent to .
Suppose that |= . We then have Mods() Mods(), leading to the
picture in Figure 1.2. It should be clear now that Mods()Mods() = Mods()
and, hence, Mods( ) = Mods() by the properties of Mods. It then follows
immediately that is equivalent to by definition of equivalence.

1.5 Reductions among logical relationships and

properties
One can state a number of reductions between logical properties and relation-
ships, some of which are shown in Table 1.2. Specifically, this table shows how
the relationships of implication, equivalence, mutual exclusiveness, and exhaus-
tiveness can all be defined in terms of the properties of satisfiability and validity.
These reductions are useful in practise. Suppose for example that we have an
algorithm for testing satisfiability. We can then immediately use this algorithm
to test whether implies by testing whether is satisfiable.
 c 2004
Adnan Darwiche 11

Mods ( )

Mods( )

Figure 1.3: The relationship between the models of arbitrary sentence and .
The shaded area represents the set of worlds Mods() Mods().

One can also prove these reductions using the Mods function and its prop-
erties. To illustrate this technique, we will next prove two of the reductions in
Table 1.2, which are known as the Refutation and Deduction theorems.

1.5.1 The Refutation Theorem

The Refutation Theorem states that:

implies , |= , iff is inconsistent.

This theorem is quite commonly used in mathematical proofs under the label
proof by contradiction. That is, to show that implies , we assume ,
assume , and then try to show a contradiction. When one is found, we
declare that implies . This method of reducing the implication test to a
satisfiability test is also predominant in automated reasoning systems as we
shall discuss later.
To prove this theorem we recall that

|= iff Mods() Mods(), and

is inconsistent iff Mods( ) = .

It suffices then to show that Mods() Mods() iff Mods( ) = .

Consider now Figure 1.3 which depicts the relationship between Mods()
and Mods() for arbitrary sentences and . It should be clear from this figure
that Mods() Mods() iff the set Mods() Mods() = . By properties
of the Mods function, we have Mods() Mods() = Mods( ). Hence,
Mods() Mods() iff Mods( ) = , which proves the theorem.

1.5.2 The Deduction Theorem

The Deduction Theorem states that:

implies , |= , iff is valid.

12 Class Notes for CS264A, UCLA

We will now prove this theorem assuming that we have accepted the Refutation
theorem, and by showing that is valid iff is inconsistent.
First, we recall that sentence is valid iff its negation is inconsistent:

Mods() = iff Mods() = .

This follows immediately from the properties of Mods since Mods() = Mods().
All we have to show now is that is equivalent to the negation of ,
which proves that is valid iff is inconsistent:

() is equivalent to by de Morgans law and double negation.

is equivalent to by definition of implication.

This completes the proof.

1.6 Quantified propositional logic

One can define a few more operators on propositional logic sentences, which
enrich the language and lead to what is known as quantified propositional logic.

Conditioning
The first of these operators is known as conditioning and allows one to eliminate
a variable from a knowledge base, given that we know the value of that variables.
Specifically, to condition a knowledge base on a positive literal P is to replace
every occurrence of P in by true. Moreover, to condition a knowledge base
on a negative literal P is to replace every occurrence of P in by false.
The result of conditioning knowledge base on literal is denoted by | and
a key property of | is that it does not mention the variable of literal .
Using the conditioning operator, one can define Booles expansion which
has major applications in recursive algorithms that we discuss later. Given a
knowledge base and a variable P , Booles expansion allows us to rewrite
as follows:
P |P P |P .
The key property of this expansion is that both |P and |P do not mention
variable P . Therefore, this expansion allows us to reduce a query with respect
to a knowledge base over n variables into two queries on knowledge bases over
n 1 variables. If Booles expansion is applied recursively to the newly created
knowledge bases, it can then be used to decompose a complex knowledge base
down to trivial knowledge bases that are equivalent to either true or false. We
will see many examples of such application in future chapters.
If the value of a variable is not known, we can still eliminate it from a
knowledge base while preserving important properties of the original knowledge
base. There are two methods for eliminating a variable in this case, which we
discuss next.
 c 2004
Adnan Darwiche 13

Existential quantification
The first operator for eliminating a variable P from a knowledge given
that the value of P is unknownis to existentially quantify variable P out of
knowledge base :
def
P = |P |P .
Consider the knowledge base

A B
=
B C,
which can be written as

A B
=
B C.
We then have
|B = (A true) (true C) = C,
and
|B = (A false) (false C) = A,
leading to
B = A C = AC.
Note here that B is implied by . This is true in general for any knowledge
base and any variable:
|= P .
We even have a stronger property for existential quantification. Let be the set
of all propositional variables. Then not only is P implied by , but it also
contains all the information that contains about the variables \ P . That is,
if is a sentence which is constructed using the variables \ P , then

|= iff P |= .

Therefore, one can interpret P as the strongest sentence over variables \ P

that is implied by . This is why P is usually referred to as the forgetting
of variable P from , as it represents the result of erasing what says about
variable P (without affecting what says about other variables). This is also
why P is sometimes referred to as the projection of on variables \ P , as
it represents all that says above variables \ P .
Consider now the circuit in Figure 1.5, and the following knowledge base
describing its behavior:


A X



A X

AB Y
=

(A B) Y



X Y C

(X Y ) C.
14 Class Notes for CS264A, UCLA

Using the definition of existential quantification, one can verify that

A B C
XY =
A B C,

which corresponds to a less informative description of the circuit behavior. Fi-

nally, one can verify that

XY |C = A B,

which is even a less informative description of the given circuit as it only char-
acterizes circuit input under which the circuit will produce a high output.

Universal quantification
The other method for eliminating a variable P from a propositional knowledge
base is by universally quantifying P out of knowledge base :

def
P = |P |P .

Universal quantification satisfies properties which are dual to those satisfied by

existential quantification:
P |= .

Moreover, if is a sentence constructed from variables \ P , then

|= iff |= P .

Therefore, one can interpret P as the weakest sentence over variables \ P

that implies .

Duality
Existential and universal quantifiers are related by the following dualities:

P = (P ),

P = (P ).

It is always possible to quantify over more than one variable and the order of
quantification does not matter for the same quantifier. Hence, P Q is equiv-
alent to QP and similarly for universal quantification. We will therefore
simply write P, Q and P, Q instead of P Q and P Q, respectively,
whenever convenient.
 c 2004
Adnan Darwiche 15

Semantics of quantified propositional logic

We have defined conditioning, existential quantification, and universal quantifi-
cation as transformations on classical propositional sentences. Another way of
defining these operators is by extending the satisfaction relationship between
worlds and propositional sentences to cover sentences that include these opera-
tors. Specifically, let [P ] denote a world which agrees with on every variable
except possibly on P , where [P ] sets P to true. Similarly, let [P ] denote a
world which agrees with on every variable except possibly on P , where [P ]
sets P to false. We then have:

|= |P iff [P ] |= .

|= |P iff [P ] |= .

|= P iff [P ] |= or [P ] |= .
|= P iff [P ] |= and [P ] |= .

Given the above extensions to the satisfiability relationship between worlds and
sentences, one can immediately extend the function Mods to sentences that
include conditioning and quantification operators. All logical properties and re-
lationships are then defined, therefore, providing formal semantics to quantified
propositional logic.

1.7 Queries and transformations on KBs

We have defined the syntax and semantics of propositional logic in previous
sections, which allowed us to formally define the representational aspects of
our approach to automated reasoning. We will identify in this section a class
of operations on knowledge bases that will constitute the building blocks from
which we will compose more complex operations as we address applications in
future chapters.
We will divide our operations into two classes: queries and transformations.
Queries are operations that return information about a knowledge base, while
transformations are operations that modify a knowledge base.

Queries
The set of queries we have identified are as follows, where and are knowledge
bases:
CO Testing whether a is satisfiable.

VA Testing whether a is valid.

CT Counting the models of .

EN Enumerating the models of .

 16 Class Notes for CS264A, UCLA

IP Testing whether implies a disjunction of literals (clause) .

PI Testing whether a conjunction of literals (term) implies .

IM Testing whether implies .

EQ Testing whether is equivalent to .

Probably the key observation about the above queries is that their difficulty
depends on the syntactic form of given knowledge bases. For example, CO is
hard on arbitrary propositional sentences, but becomes easy if the sentence is
a disjunction of terms, known as Disjunctive Normal Form (DNF). Moreover,
validity is also quite difficult on arbitrary sentences, but becomes easy if the
sentence is a conjunction of clauses, known as Conjunctive Normal Form (CNF).
Another observation about the above queries is that some are more general
than others. For example, enumerating models is more general than counting
these models as one can count models as a side effect of enumeration. Moreover,
IM is more general than IP since clauses are a special case of knowledge bases.
We make this distinction, however, since IM is much more difficult than IP,
yet IP is quite useful in many applications.

Transformations
We now turn to transformations of knowledge bases, where and are knowl-
edge bases:

NEG Negating .

CON Conjoining and .

DIS Disjoining and .

CND Conditioning on a literal.

EQT Existentially quantifying variable P out of .

FQT Universally quantifying variable P out of .

Given the definition of these transformations, it may appear that they are easy
to perform as they all can be implemented in time linear in the size of given
KBs. This ceases to be true, however, once we restrict the syntactic form of
KBs. For example, negating a KB is straightforward, but becomes difficult if
we insist that both the original and negated KB are in CNF. We will in later
chapters discuss a large number of syntactic forms and enumerate the difficulty
of the above operations on each of these forms.
 c 2004
Adnan Darwiche 17

Figure 1.4: A simple circuit.

Putting the operations to use

To provide an example of how these queries and transformations can be put to
use in implementing reasoning applications, let us consider a simple circuit with
two cascaded inverters as given in Figure 1.4. The knowledge base for this
circuit is shown below:


ok (X) A B,

ok (X) A B,
= .

ok (Y ) B C,

ok (Y ) B C

Suppose now that we observe the input A to be high, the output C to be

low, and we are interested in knowing whether this is a normal device behavior.
This can be decided by asking whether the extended KB,

{A, C, ok (X), ok (Y )}

is satisfiable. The answer is no in this case, indicating that the observed be-
havior is abnormal. Hence, we used the satisfiability query to decide whether a
particular device behavior is abnormal.
Let us now define a device state to be an assignment of true/false to the
health variables ok (X) and ok (Y ). And suppose that our goal is to count the
number of states that the circuit can be in, given that we have observed the
input A to be high, and the output C to be low. One way to do this is using
two more operations on the KB: existential quantification and model counting.
Specifically, we existentially quantify variables A, B and C out of the KB
{A, C}, leading to the new KB, {ok (X) ok (Y )}, which characterizes the
circuit health under given observation. We then count the models of this new
18 Class Notes for CS264A, UCLA

KB, which turns out to be 3. These models correspond to the possible states
of given circuit and we may want to enumerate them, leading to the following:
{ok (X), ok (Y )}, {ok (X), ok (Y )}, and {ok (X), ok (Y )}.

1.8 Boolean functions

The operations we defined in the previous section on propositional knowledge
bases correspond to similar operations on Boolean functions. In fact, some of
the more important results on these operations have originated in the literature
on Boolean functions and then translated into the language of propositional
logic. Understanding the connection between propositional knowledge bases
and Boolean functions is therefore important for a comprehensive coverage of
the literature on propositional reasoning. Moreover, some of the notions that
exist in both literatures are more intuitively expressed in one representation
than in the other. Hence, our ability to freely navigate between the two repre-
sentations can aid our comprehension of such notions. Our goal in this section
is then to provide an introduction to Boolean functions and how they relate to
propositional knowledge bases.

1.8.1 Basics of Boolean functions

A Boolean variable is one which takes values in 1, 0. A Boolean function f over
Boolean variables X1 , . . . , Xn is a function that maps each value of these vari-
ables into 1 or 0. Boolean functions can be constructed using three operations
on Boolean values:

If b is a Boolean value, then b is the complement of that value: b = 1 b.

If b1 and b2 are Boolean values, then b1 .b2 is the product of these values,
and b1 + b2 is the maximum of these values.

Using these operations, one can build complex Boolean functions from simpler
ones. For example, if f (X1 , . . . , Xn ) and g(X1 , . . . , Xn ) are two Boolean func-
tions, then f (X1 , . . . , Xn ), f (X1 , . . . , Xn ).g(X1 , . . . , Xn ), and f (X1 , . . . , Xn ) +
g(X1 , . . . , Xn ) are Boolean functions.
The conditioning and quantification operators have parallels on Boolean
functions. Specifically, the conditioning operator corresponds to the restriction
operator, where the restriction of function f (X1 , . . . , Xn ) to Xi = b, denoted
f |Xi =b , is the function f (X1 , . . . , Xi1 , b, Xi+1 , . . . , Xn ), which results from fix-
ing the value of variable Xi to b. Note that the restriction f |Xi =b is a Boolean
function over n 1 variables.
The Shannon expansion of Boolean function f (X1 , . . . , Xn ) with respect to
variable Xi is given by the following identity:

f = Xi .f |Xi =1 + Xi .f |Xi =0 .
 c 2004
Adnan Darwiche 19

This expansion allows one to express a Boolean function over n variables in

terms of two Boolean functions, each over n 1 variables. This expansion
parallels Booles expansion for propositional logic.
One can existentially quantify over variables in a Boolean function:
def
Xi f = f |Xi =1 + f |Xi =0 .

The expression Xi f (X1 , . . . , Xn ) is then a Boolean function over only n 1

variables, X1 , . . . , Xi1 , Xi+1 , . . . , Xn .
A solution to a Boolean function f (X1 , . . . , Xn ) is an assignment of values
to variables X1 , . . . , Xn under which the function f evaluates to 1. Suppose
now that g is the result of existentially quantifying variable Xi in function f ,
g = Xi f , and let be a solution to function g. There are two ways in which
the solution can be extended: by assigning value 1 to Xi , leading to + , or by
assigning value 0 to Xi , leading to . By construction of g, we are guaranteed
that either + or must be a solution to function f . That is, some extension
of a solution to function g is guaranteed to be a solution to function f .
One can also universally quantify over variables in a Boolean function:
def
Xi f = f |Xi =1 .f |Xi =0 .

The expression Xi f (X1 , . . . , Xn ) is also a Boolean function over only n

1 variables, X1 , . . . , Xi1 , Xi+1 , . . . , Xn . Suppose now that g is the result of
universally quantifying variable Xi in function f , g = Xi f ; let be a solution
to function g; and let + and be its two extensions with respect to variable
Xi . In this case, we are guaranteed that both + and are solutions to function
f . That is, every extension of a solution to function g is guaranteed to be a
solution to function f .
One can define notions such as satisfiability, validity, and implication on
Boolean functions. For example, a Boolean function f is satisfiable, written
f 6= 0, iff it has at least one solution; and f is valid, written f = 1, iff every
assignment of values to its variables is a solution. A Boolean function f implies
another Boolean function g, written f g, iff every solution of f is also a
solution of g. Other logical notions can be defined similarly.

1.8.2 Boolean functions of digital circuits

Consider the digital circuit in Figure 1.5. One can define at least three types
of Boolean functions with respect to this circuit (and any other similar circuit),
that have different information contents.
A Type III Boolean function for a circuit is a function over the circuit inputs.
There is one Type III function fO for each circuit output O. The function fO
evaluates to 1 under a particular assignment of values to circuit inputs iff the
circuit output O assumes value 1 under that particular assignment. Since the
circuit in Figure 1.5 has only one output, it then has only one Type III Boolean
function:
20 Class Notes for CS264A, UCLA

A B

X Y

Figure 1.5: A digital circuit.

A B fC (A, B)
0 0 1
0 1 1
1 0 0
1 1 1

A Type II Boolean function for a circuit is a function over the circuit inputs
and outputs. When evaluated under a particular assignment of values to inputs
and outputs, the Boolean function returns 1 iff the circuit generates the given
outputs under the given inputs. The Type II Boolean function for the circuit
in Figure 1.5 is:

A B C g(A, B, C)
0 0 0 0
0 0 1 1
0 1 0 0
0 1 1 1
1 0 0 1
1 0 1 0
1 1 0 0
1 1 1 1

A Type I Boolean function for a circuit is a function over the circuit inputs,
outputs, and all internal wires (that are neither inputs nor outputs). When
evaluated under a particular assignment of values to wires, the Boolean function
returns 1 iff the given assignment is compatible with the circuit behavior. The
Type I Boolean function for the circuit in Figure 1.5 is a function h over variables
A, B, C, X, Y . It is shown partially below:
 c 2004
Adnan Darwiche 21

A B C X Y h(A, B, C, X, Y )
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 1 0
0 0 0 0 1 0
.. .. .. .. .. ..
. . . . . .
0 0 1 1 0 1
.. .. .. .. .. ..
. . . . . .
We can always obtain Type III and Type II functions from a Type I function
using existential quantification. For the above circuit, we have g = XY h.
Moreover, we have fC = XY h|C=1 . The Type III function fC can also be
obtained from the Type II function using fC = g|C=1 . These constructions are
general and work for other circuits as well.

1.8.3 Knowledge bases and Boolean functions

There is a onetoone correspondence between Boolean functions and equiva-
lence classes of propositional knowledge bases. Consider the Boolean function
f (X1 , . . . , Xn ). A propositional knowledge base over variables X1 , . . . , Xn
corresponds to Boolean function f iff the models of correspond to the solu-
tions of f . We also say in this case that f is the characteristic function of .
Note that there are many propositional knowledge bases that correspond to a
particular Boolean function, but all these knowledge bases must be equivalent.
Consider the following Boolean function f over variables A, B:

A B f (i )
1 : 1 1 0
2 : 1 0 1
3 : 0 1 1
4 : 0 0 1
The solutions of this function are 2 , 3 , 4 . Consider now the knowledge base
= (A B) and its corresponding set of possible worlds:

world A B
1 true true
2 true false
3 false true
4 false false
The knowledge base corresponds to the Boolean function f since Mods() =
2 , 3 , 4 . The knowledge base = A B also corresponds to function f
since is equivalent to .
The language of propositional logic can therefore be viewed as a specification
language for Boolean functions, since each knowledge base corresponds to a
22 Class Notes for CS264A, UCLA

unique Boolean function which we shall denote by f . Moreover, the tools for
reasoning within propositional logic can be viewed as an apparatus for reasoning
about Boolean functions.

1.9 Propositional logic with discrete variables

Standard propositional logic assumes that propositional variables have Boolean
values true/false. Hence, the statement Pi can be interpreted as a shorthand for
Pi = true. It is possible to generalize propositional logic by allowing discrete
variables that can take any finite number of values, not just Boolean values.
For example, we may extend the earthquakeburglaryalarm example so that
variable Alarm is discrete and has three values off, low and high. In this case,
the simplest statements one can write in the logic would be of the form Pi =
v, indicating that variable Pi takes on value v. The new syntax is a simple
extension of what we have given before:
If Pi is a variable and v is one of its values, then Pi = v is a sentence.
If and are sentences, then , , and are also sentences.
We can also use Pi as a sentence, assuming that variable Pi has values true/false,
in which case the sentence is interpreted as a shorthand for Pi = true. Going
back to the earthquakeburglaryalarm example, and assuming that variables
Earthquake and Burglary are Boolean, one may write the following sentences:

Earthquake Alarm = low

Burglary Alarm = high
Earthquake Burglary Alarm = off.

The semantics of this extended logic in based on a more general notion of a

world, which is an assignment of values (not necessarily Boolean) to variables.
Continuing with the earthquakeburglaryalarm example, we have twelve pos-
sible worlds:

world Earthquake Burglary Alarm

1 true true high
2 true true low
3 true true off
4 true false high
5 true false low
6 true false off
7 false true high
8 false true low
9 false true off
10 false false high
11 false false low
12 false false off
 c 2004
Adnan Darwiche 23

One can also define [Pi=v] to be a world that agrees with on all variables,
except possibly on P where [Pi =v] assigns the value v to Pi . We can then
extend the satisfiability relationship |= as follows:
|= Pi = v iff (Pi ) = v.

|= iff 6|= .
|= iff |= or |= .
|= iff |= and |= .
|= |(Pi=v) iff [Pi=v] |= .
|= P iff [Pi=v] |= for some value v.
|= P iff [Pi=v] |= for all value v.
Based on this relationship, other logical notions such as satisfiability, validity,
implication and equivalence can be defined, therefore, providing formal seman-
tics to propositional logic with discrete variables.