A BRIEF INTRODUCTION TO MODAL LOGIC

JOEL MCCANCE
Abstract. Modal logic extends classical logic with the ability to express not
only P is true, but also statements like P is known or P is necessarily true.
We will define several varieties of modal logic, providing both their semantics
and their axiomatic proof systems, and prove their standard soundness and
completeness theorems.

Introduction
Consider the statement It is autumn, thinking in particular of the ways in
which we might intend its truth or falsity. Is it necessarily autumn? Is it known
that it is autumn? Is it believed that it is autumn? Is it autumn now, or will it
be autumn in the future? If I fly to Bombay, will it still be autumn? All these
modifications of our initial assertion are called by logicians modalities, indicating
the mode in which the statement is said to be true. These are not easily handled by the truth tables and propositional variables of the propositional calculus
(PC) taught in introductory logic and proof courses, so logicians have developed
an augmented form called modal logic. This provides mathematicians, computer
scientists, and philosophers with the symbols and semantics needed to allow for
rigorous proofs involving modalities, something that until recently was thought to
be at best pointless, and at worst impossible.
This paper will provide an introductory discussion of the subject that should be
clear to those with a basic grasp of PC. We will begin by discussing the history
and motivations of modal logic, including a few examples of such logics. We will
then introduce the definitions and concepts needed for a rigorous discussion of the
subject. The third section will show that these systems behave themselves that
is, the set of provable statements and the set of true statements are identical. We
will conclude with a discussion of some of the technical complexity of introducing
quantification into modal logic.
1. History and Motivations
As with its classical cousin, the modern interest in modal logic begins with
Aristotle.1 In addition to his syllogisms dealing with categorical statements, the
Greek thinker wished to formalize the logical relationships between what is, what is
necessary, and what is possible. Unfortunately his treatment of modality suffered
from a number of flaws and confusions, and while his categorical syllogisms became
a staple of classical education, modal logic was dismissed as a failure. Kant and
Frege both argued that modalities added no pertinent information to an argument,
The author would like to thank Professor Jen Brown for her helpful comments and Professor
Bob Milnikel for his advice and support.
1
This historical information in this section is drawn from Fitting [1].
1

JOEL MCCANCE

merely hinting at why we might believe a given statement to be true. They believed
that no more or less could be derived from the modal form a statement P that from
P itself.
This claim has come to be seen as false. After all, if two statements are equivalent, they ought to imply each other. It seems reasonable to say that if P is the case
then P must be a possible state of affairs, since what is true cannot be impossible.
However, it is quite a bit less obvious to say that because P is possible, P is the
case. It is possible, for example, that my bearded friend Charles is actually a very
hirsute woman, yet there is no reason to believe that this is actually true. So while
actuality implies possibility, possibility does not imply actuality. It seems, then,
that there is more to modality than Kant and Frege suspected; modal statements
are not quite equivalent to their nonmodal counterparts.
With the growing acceptance of such arguments in the past fifty years or so,
there has been a revival of interest in modal logic, the product of which has been
a number of interesting new modalities. Temporal logic, for instance, considers
whether P is true now, will be true at some point in the future, or has been true
in the past. This is actually a multimodal logic since it uses a number of modes
of truth within a single language. Epistemic logic the logic of knowledge and
knowing can also be multimodal. The modality in this case is that of whether
agent A knows P or, alternatively, whether P could be true given what A knows.
Related is the interpretation of  which reads P as P is provable, a reading that
has great use in the field of mathematical logic.
A final, interesting twist on the theme is action-based logics. As the name would
imply, what we are interested in here is how actions alter the truth of a statement.
The example in the introduction of flying to Bombay would be an action modality.
Say It is autumn where I am is true and I am in Ohio is true. Then after flying
to Bombay, both these statements will become false, as it cannot be autumn on
both sides of the world simultaneously and, of course, I will no longer be in Ohio.
While trivial in this example, a multimodal action-based logic provides a means of
teasing out the finer implications of complex sequences of actions.
2. Propositional Modal Logic
Any complete system of logic needs at least three components: a rigorous language for writing out the statements in question, a means of interpreting the statements and determining their truth value, and a means of writing proofs.2
2.1. The Language of Modal Logic. Any language, be it English, Chinese, or
the language of logic, must have symbols and rules for combining them. In a
spoken language the symbols are words and the rules are grammar. Our case is
analogous, although rather than inventing a new system from whole cloth, modal
logic begins with the familiar language of PC and just adds two new operators to
handle modality. The result is the following set of symbols:
A countably infinite set of letters A, X, P1 , P2 , &c., called propositional
variables;
The unary operators , , and ;
The binary operators =, , and ; and
Brackets (, and ).
2The definitions in this section are adapted from Hughes [3] and Fitting [1].

A BRIEF INTRODUCTION TO MODAL LOGIC

P
T
F

P P
T
T

Figure 1. Truth table for P P .

We read , , , , , and = as box, diamond, not, or, and, and
implies.  and are, of course, our new, modal operators. Their generic names
allow us to create a single, rigorous system that can be adapted for the various
modalities we may wish to implement. All these names are, of course, merely
names thus far. We have not yet bestowed on them any formal interpretation.
The next step, then, is to define which combinations of these symbols are legal
which strings make sense given the interpretation we would like to eventually
give. These rules reflect the intuition that tells us that (P Q) is a good formula,
yet ) P Q is symbolic gibberish. These good formulae we will call well-formed
formulae, or wffs for short.
Definition 2.1. A well-formed formula or wff is any formula that meets one of
the following three rules.
Any propositional variable P standing alone is a wff.
When is a wff, so are (), (), and ().
When and are both wffs, so are ( ), ( ), and ( = ).3

Note the recursion of this definition, first defining a base case (called the atomic
formulae) and going on to define more complex formula in terms of those wffs
already known to us. This will give us a powerful tool for future proofs where
we first prove something about isolated propositional variables and go on to show
that it holds for the second and third formation rules listed above. This is called
induction on the complexity of a formula, and will play a crucial role in our proofs
for soundness and completeness below.

2.2. Determination of Truth. In PC we are primarily interested in the tautologies. These are the formulae that are true no matter what the truth of the
propositions involved, things like (P Q) = P and P P . These are called
the valid formulae of PC and are determined by simply checking the truth table
for the formula in question. For example, we can tell that P P is valid because
every line of truth table in Figure 1 comes out to T.
The situation for modal logic is somewhat more complex. After all, the whole
point of our new symbols is to indicate a judgement that is independent of the truth
of the formula in question. Nonetheless, we want an extension of the same intuitive
notion: that the valid formulae are those that are True no matter what or True
in all situations. The difference is that now it is possible that some propositions,
while not tautological, will nonetheless always evaluate true. For example, it has
been a popular move in theology to claim that it is necessary that there exist a
greatest conceivable being. Such theologians are not generally claiming that Gods
existence is a tautology, but rather that in every conceivable world the proposition
God exists is true. Therefore, they argue, it is a necessary truth that God exists.
3The parentheses in this and the preceding rule, while crucial in some instances for rigor, will
often be omitted when the meaning is clear from context and convention. This way we will not
have to write ((P ) Q) when P Q is perfectly understandable.

JOEL MCCANCE

It is this reading of necessarily true as true in all possible worlds that lead
to the most popular interpretation of modal logic: Kripkes many-world semantics.
Under this interpretation, the truth of a statement is relative to the world in question. For propositional formulae, this is determined simply by examining the state
of affairs in that world. So if P and Q are both true in the current world, P Q
will be true in this world. The more interesting case comes with our new operators,
 and . P is defined to be true in a world whenever P is true in all accessible
worlds. How we define accessibility depends on the modality, but conceivable is a
common one for the necessary/possible modality. So if P is true in all conceivable
worlds, P is true that is, P is necessarily true. P is similar, although in this
case the modality is that of possibility. If P is true in at least one accessible world,
P will be true as well since it is true somewhere, it must not be impossible.
To consider another example, say we were using P to mean X believes P ,
where X is some person or ideology. The possible worlds here are not really worlds
at all, but people, ideologies, institutions anything that can be said to believe
a proposition. Accessibility in this case is interpreted as trusts. So if Platonists
trust physics, then physics is accessible to Platonism. More elaborately, say that
Russell is a node in this network of people and ideologies. For this example, say
Russell trusts physics, Richard Rorty, and atheism exclusively. Then P is only
true for Russell if P is true for physics, Richard Rorty, and atheism. If P is not
true in any one of these, than Russell does not believe P . Similarly, if only Richard
Rorty believes P , then P is true for Russell. He can see how one would believe
P , but is not fully convinced. Finally, note that the truth or falsehood of P in
Scientology will have no effect on what Russell believes because he does not trust
it Scientology is not accessible to Russell by the relation trusts.
With these examples in mind, we can now rigorously define the semantics of our
symbols.
Definition 2.2. Let W be a non-empty set of what we will call possible worlds.
Let R be a binary relation from W to W , which we will call an accessibility relation.
Together, hW, Ri form a frame.

Definition 2.3. Let hW, Ri be a frame and let (read forces) be a binary relation
between W and the set of all well-formed formulae. Let W . We will assume
that obeys the following rules
For all propositional variables P , either P or P .
If is a wff, then if and only if 6 .
If and are wffs, then ( ) if and only if or .
If and are wffs, then ( ) if and only if and .
If and are wffs, then ( = ) if and only if 6 or .
 only if for every W , R implies that .
only if there exists a W such that R and P .
Together, hW, R, i is a propositional modal model, which we will generally shorten
to model.
There are a few items of note in this definition. First, observe that all but the
last two cases are identical to the truth-table semantics of PC. Only the final cases
add anything new. It is also worth observing that just as we can rewrite and
using just negation and implication, we can also rewrite in terms of negation and
. Think of what it would mean for P to be false in a world. This would mean

A BRIEF INTRODUCTION TO MODAL LOGIC

Logic Frame Conditions

K
no conditions
A BRIEF INTRODUCTION
TO MODAL LOGIC
D
serial4
Frame Conditions
T Logic
reflexive
no conditions
B Kreflexive,
symmetric
D
serial4
K4 Ttransitive
reflexive
reflexive,transitive
symmetric
S4 Breflexive,
transitivesymmetric, transitive
S5 K4reflexive,
S4
S5

reflexive, transitive
reflexive, symmetric, transitive

Figure 2. Common systems of modal logic and the restrictions

Figure
2. Common
systems
of modal
logic and the restrictions
they place
on the
relations
between
frames.
they place on the relations between frames.
B
T

K4

S5

S4

Figure 3. Relative strength of some standard frames in modal logic.

Figure 3. Relative strength of some standard frames in modal logic.

that in every accessible world P is false, so the negation of P will be true in all

possible
worlds world
that is, !P
true. So
is equivalent
!P
A helpful
that in every
accessible
P isisfalse,
soPthe
negationto of
P .will
be true in all
analogy here is to think of why (x)(P (x)) is equivalent to (x(P (x))). To say
possible worlds
isthat
P
equivalent
tofor
P
. A helpful
that there
an xis,
such
that Pis(x)true.
is trueSo
is toP
say is
that
it is false that
every x,
P (x)isisto
false.
Similarly,
X (x)(P
simply asserts
it is false thattoX(x(P
is false in every
analogy here
think
of why
(x)) that
is equivalent
(x))). To say
that there accessible
is an x world.
such that P (x) is true is to say that it is false that for every x,
Now that we have a clear interpretation of our symbols, we can finally define
P (x) is false.
simply
it is false that X is false in every
whatSimilarly,
it means for aX
formula
to be asserts
valid in a that
given system.
accessible world.
Definition 2.4 (L-valid). Let #W, R, #$ be a model. We say that this model
Now that
we have
clear#W,
interpretation
our symbols,
define
is based
on thea frame
R$. Let be a of
well-formed
formula. we
iscan
validfinally
in
#W, R,for
#$ a
if formula
# for every
valid
W . is
in the system.
frame #W, R$ if it is valid in
what it means
to be
invalid
a given
every model based on #W, R$. Finally, if is valid in a collection of frames L, then

say it(L-valid).
is L-valid.
Definitionwe2.4
Let hW, R, i be a model. We say that this model
is based on Since
the frames
framearehW,
Leta relation,
be ait well-formed
formula.
is valid in
justRi.
a set with
is natural to want
to choose collecof frames
R. valid
This is exactly
logicians
have done,
hW, R, i tions
if
for based
everyonproperties
W . of is
in thewhat
frame
hW, Ri
if it is valid in
beginning with the system K, which places no restrictions on the frame whatsoever.
every model
based on hW, Ri. Finally, if is valid in a collection of frames L, then
Other common systems and frame conditions are listed in Figure 2.
we say it is L-valid.
Note the interconnections implied by this table. For example, any formula that is
K-valid ought to be valid in all these systems, since its truth relies on no particular

Since frames
are justSimilarly,
a set with
it is
want
to choose
collecframe structure.
whataisrelation,
true in B will
be natural
true in S5,to
since
the former
is
identical
to theon
latter
with the exception
of notisnecessarily
Thehave done,
tions of frames
based
properties
of R. This
exactly being
whattransitive.
logicians
exact relationships are illustrated in Figure 3.
beginning with
the system K, which places no restrictions on the frame whatsoever.
Other common
frame
conditions
are
listed
inis Figure
2..
4For systems
every world and
W there
exists
at least one
W such
that
accessible to
Note the interconnections implied by this table. For example, any formula that is
K-valid ought to be valid in all these systems, since its truth relies on no particular
frame structure. Similarly, what is true in B will be true in S5, since the former is
identical to the latter with the exception of not necessarily being transitive. The
exact relationships are illustrated in Figure 3.
2.3. An Axiomatic Treatment of Proofs. We now have a means of writing
statements and of talking about which are true in which circumstances. However,
4For every world W there exists at least one W such that is accessible to .

JOEL MCCANCE

the fact that the reader has probably taken an introductory proof course should
suggest that more is required from a system of logic. We do not want to have
to compute truth tables for every possible world whenever we want to establish
the veracity of a statement. Instead we want a system that formalizes the usual
reasoning process of mathematics: beginning from what we know and applying
rules and axioms to generate new theorems.
In the discussion that follows we will consider the system K. Since K makes no
assumptions about the structure of the frames, we will be able to augment K with
more axioms to transform it into any of the systems discussed so far.
First, the definition of a proof:
Definition 2.5. An axiomatic proof is a finite sequence of formulae, each of which
is either an axiom or else follows from the earlier terms of the sequence by one of
the rules of inference. An axiomatic theorem is the last line of a proof.
The axioms of our system are as follows:
Definition 2.6. There are two classes of axioms for K
Classical Tautologies All valid formula of PC (that is, all the tautologies of
traditional propositional logic) will be taken as axioms.
Schema K For any wffs and , we will assume that
( = ) = ( = ).
Our proof system will also have two rules of inference.
Definition 2.7. A rule of inference is an ordered pair (, ), where is a set of
wffs and is a single wff. If the propositions of are theorems of the system, so is
.
K has two rules of inference.
Modus ponens: ({, = }, ).
Necessitation: ({}, ).
Since we are introducing axioms, it is worth taking a moment to ensure that
they are reasonable. While the classical tautologies and modus ponens are the
same as in PC, Schema K and Necessitation may take some argument. We will
start with Schema K: (X = Y ) = (X = Y ). Say that we have a proof
for (X = Y ) and that we are interpreting  in terms of necessity. Then what
we are saying is that we have proven it necessarily true that X = Y . If we can
prove that X is necessarily true then we will have shown that X is true in every
possible world. Since X = Y must be true in every possible world, it is reasonable
to say that Y must be as well and that Y is therefore necessarily true. Schema K
is also reasonable under epistemic logic. If know that X = Y and we know X,
we also know Y .
First, note that since we are in a proof system, X does not mean X is the case.
We are rather asserting that X is provable. Consider the two unimodal logics we
have discussed: the logic of necessity and epistemic logic. For the necessary/possible
modality we are claiming that if we have a proof for X from our other axioms and
rules of inference, X must be necessarily true. This is exactly the sort of result we
want from a proof system. Similarly, in epistemic logic we are claiming that if we
have a proof for X, we know X. This also makes sense. If we have proven X, we

A BRIEF INTRODUCTION TO MODAL LOGIC

Name
D
T
4
B

Scheme
P = P
P = P
P = P
P = P

Figure 4. Common axiom schemes for modal proof systems.

expect X to be both true and justified. So Necessitation seems a reasonable axiom
to adopt.
Before giving an example of an axiomatic proof, we must first introduce the
Derived Rule of Regularity. Derived rules are rules of inference that can be deduced
from the axioms and rules of inference already available. We can treat them as
rules of inference when convenient, since they can always be translated back into
the official axioms and rules when need be.
Definition 2.8 (Derived Rule of Regularity). ({X = Y }, X = Y .
This is a derived rule since for any X and Y , the following official steps will
produce the same result:
Occurs somewhere in the proof
X = Y
(X = Y )
Necessitation on the previous
(X = Y ) = (X = Y ) Axiom K
X = Y
Modus Ponens on the previous two lines
Example 2.9 (An axiomatic proof). (X Y ) = (X Y ).
The proof is as follows:
(X Y ) = X
(X Y ) = X
(X Y ) = Y
(X Y ) = Y
[(X Y ) = X] =
{[(X Y ) = Y ] =
[(X Y ) = (X Y )]}
6 [(X Y ) = Y ] =
[(X Y ) = (X Y )]
7 (X Y ) = (X Y )
Q.E.D. 

Proof.
1
2
3
4
5

Tautology
Regularity on 1
Tautology
Regularity on 3
Tautology
Modus Ponens, 2, 5
Modus Ponens, 4, 6

As stated before, the other systems we have introduced are identical to K, with
only a few added restrictions on the frames. Axiomatically, these systems simply
add more axiom schemes. The added axioms are show in Figure 4. The common
systems of modal logic are formed by adding combinations of the schemes to the
axiom system K, as shown in Figure 5.
3. Soundness and Completeness
It is an interesting fact that our system of proof never introduced a formal
link to our method for determining validity. How do we know, then, whether the
theorems that we prove are valid in our collection of frames? How do we know if
our proof system is strong enough to derive all the valid statements in our collection

JOEL MCCANCE

Logic
D
T
K4
B
S4
S5

Added Axioms
D
T
4
T, B
T, 4
T, 4, B

Figure 5. Common logics and the axiom schemes necessary to

properly axiomatize them.
of frames? These conditions, known as soundness and completeness respectively,
are important if a proof system is to be of any use. This section will show that the
axioms we have introduced are powerful enough to meet these conditions.
Since we know that we can write all our other operators in terms of , , =,
and , we will restrict our language to these operators in this section. While
having the full library of operators is notationally efficient and important to know,
the extra cases merely add clutter to the exposition that follows.
3.1. Soundness. Showing soundness is relatively simple. If our axioms are valid
and our rules of inference never derive an invalid statement from a valid one, then
clearly every provable formula will be valid.
We begin with the soundness of K.
Theorem 3.1 (K is Sound). If X has a proof using the axiom system K, then X
is K-valid.
Proof. We begin with the rules of inference
Modus Ponens. Assume that the wffs X and X = Y are K-valid. So for all
models based on all frames hW, Ri and for all W , X and X = Y .
Fix a frame hW, Ri and a world W . Since X = Y , we know that either
6 X or Y . Since we know that X, Y . Since our choice of world,
model, and frame within K were all arbitrary, Y is K-valid. So modus ponens is a
valid rule of inference.
Necessitation. Assume that X is K-valid. So for all models based on all frames
hW, R, i and for all worlds W , X. Fix a frame hW, Ri and a world W .
Consider a W such that R. Since X is true in every world, clearly X.
So X is true in every world accessible to . Ergo, X. Since our choices of
frame and world were all arbitrary, X is K-valid. So Necessitation is a valid rule
of inference as well.
All that remains is to prove our axioms valid. This is trivial for the Classical
Tautologies since our method of determining truth for non-modal statements is
identical to the method used in PC. Since the Classical Tautologies are exactly the
valid statements of PC, clearly they are valid. The only remaining task therefore
is to show the validity of Schema K.
Schema K. We wish to show that (X = Y ) = (X = Y ) is true for
all worlds in all frames. Let hW, Ri be a frame and a world in W . Assume that
(X = Y ) is true in . This means that if R then (X = Y ). We want
to show that X = Y is true in . We know this will be true if it is always the
case that either X is not true in or Y is true in . Assume that X is true in

A BRIEF INTRODUCTION TO MODAL LOGIC

. So for all W , R implies that X. Fix such a . Since (X = Y )

is true in and R, we know that X = Y . So Y must be true in ,
making Y true in . So in every world in any frame, either (X = Y ) is false
or X = Y is true. Hence the axiom is K-valid.
Since our axioms are sound and our rules of inference produce sound theorems
from other sound theorems, every provable X in the axiom system K is K-valid. 
The proofs for the soundness of the other systems are largely corollaries of the
above proof. Since they have the same rules of inference and only add axioms,
all that must be done is to show that the axiom schemes are L-valid, where L is
whichever collection of frames in question.
Theorem 3.2 (D, T, K4, B, S4, and S5 are Sound). Let L be D, T, K4, B, S4,
or S5. If X has a proof in the axiom system L, the X is L-valid.
Proof. Since K is valid and each system L is just K with some extra axiom schemes,
all we must do is show that each axiom scheme is L-valid in the appropriately
structured frame.
(D: P = P ) We wish to show P = P is true in all serial frames
that is, in all frames where every world is related to at least one other world.
Let hW, R, i be a model and a world in W . Say that P is true in . Then if
R, P is true in . We want to show that P is true in . In other words,
that there exists a accessible to such that P .
Since hW, Ri is serial, there exists at least one accessible to . Fix such a .
Since P is true in by hypothesis and R, we know that P . It follows
that P is false in , so there exists a world accessible to in which P is false.
Therefore P is not true in , so P is. Since our frame and choice of world
were arbitrary, P = P is D-valid. Since the only extra axiom of D is D,
D is sound.
(T : P = P ) Let hW, Ri be a reflexive frame that is, for all W , is
accessible to itself. Fix such a and assume that P is true in . Then in every
world accessible to , P is true. Since is accessible to itself, P must be true in .
So T is true in all reflexive models, including T. Since T was the only extra axiom
of this system, T is sound.
(4: P = P ) Let hW, Ri be a transitive frame. Let be a world in W and
assume P . Then for every accessible to , P is true in . Either is
related to another world or it is not. If it is not, P is vacuously true in . Say
that R. Then since the model is transitive, R. So since P is true in , P is
true in . Since our choice of was arbitrary, P is true in all worlds related to .
Therefore P is true in every related to , so P is true in . 4 is therefore
true in all transitive frame, including K4.
Since 4 was the only added axiom to this system, K4 is sound. Furthermore,
since S4 is both reflexive and transitive, T and 4 are both valid in S4. Since these
were the only extra axioms, S4 is sound.
(B: P = P ) Let hW, Ri be a symmetric, reflexive frame and let W .
Say that P is true in . Since the frame is reflexive, we know there exists at least
one world accessible to . Let be such a world. Since the frame is symmetric,
is accessible to . We know that P is true in , so P is true in . As this
holds for all worlds accessible to , P is true in .

10

JOEL MCCANCE

B is therefore valid in all symmetric, reflexive frames, including B. Since B is

reflexive, T also holds in B. Since T and B were the only added axioms, B is sound.
Finally, S5 simply requires that frames are reflexive, symmetric, and transitive.
S, 4, and B are therefore all valid in S5. Since these were the only added axioms,
S5 is sound. 
3.2. Completeness. The argument for completeness is significantly more complex
that the one for soundness. This section begins with some preliminary definitions
and results that work for any system L. However, the actual proof of completeness
will be for K, with the remaining systems as corollaries as before.
Definition 3.3 (Consistent). Let be an abbreviation for (P P ). A finite set
of formulae {X1 , X2 , ..., Xn } is L-consistent if (X1 X2 ... Xn ) = is not
provable using the L axiom system. An infinite set is L-consistent if every finite
subset is L-consistent.
Definition 3.4 (Maximal Consistency). A set S of formulae is maximally Lconsistent if S is L-consistent and no proper extension of it is. That is, if S S 0
and S 0 is also L-consistent, then S = S 0 .
Theorem 3.5. If S is L-consistent, then it can be extended to a maximally Lconsistent set.5
Proof. Our general strategy for this proof is to add either formulae or their negations one at a time to a consistent set until we run out of formulae. The resultant
set will be consistent because every finite subset of it was consistent. It will also
be maximal, since every formula or its negation is already in the set.
Let S be an L-consistent set of formulae. Since every formula is finite combination of a finite set of symbols, we know from set theory that the set of all
wffs is countable. We can therefore map them to the natural numbers like so:
X1 , X2 , X3 , .... We can use this fact to construct a sequence of sets S0 , S1 , S2 , S3 , ...
as follows:
S0
Sn+1

= S
(
=

Sn Xn+1
Sn

if Sn Xn+1 is consistent
otherwise.

Note first of all that by construction, each Si is consistent. Furthermore, Si

Si+1 for every i N. We can therefore define the limit of this sequence S as
S0 S1 S2 .... Our claim is that S is maximally consistent that is, S is
L-consistent and if S S 0 is consistent then S = S 0 .
S is L-consistent. We will show this by contradiction. Assume that S was
not L-consistent. Then there exists some finite subset Z = {Z1 , Z2 , Z3 , ..., Zn } of
S such that (Z1 Z2 ... Zn ) = . But we know that not every Zi came
from S since S was L-consistent. We also know that for every Zi 6 S there exists
some j N so that Zi 6 Sj Sj1 . (That is, Sj is the first set in the sequence
containing Zi .) Since the set of Zs is finite we can fix k equal to the greatest such
j. As indicated above, S0 S1 ... Sk . So since each Zi is in Sj for some j k,
it follows that each Zi Sk . Therefore Sk is inconsistent with L. But we know
5The proof of this theorem and those that follow have been adapted from Fitting [1].

A BRIEF INTRODUCTION TO MODAL LOGIC

11

that Si is L-consistent for every i N, so this gives us a contradiction. Therefore

S is L-consistent.
S is maximal. This will also be proven by contradiction. Say there was a proper
extension of S that maintained L-consistency. Then there is some formula Z 6 S
such that S {Z} is L-consistent. But we know that our list of formulae X1 , X2 , ...
contained all formulae, so there exists i N so that Z = Xi . So either Xi Si or
Xi 6 Si . If Xi Si , then Xi S and S {Xi } is not a proper extension of S .
So say that Xi 6 Si . Then it must be that Si1 Xi was not consistent. Since
Si1 S , S {Xi } would not be consistent. Therefore there cannot exist a proper
extension of S that maintains L-consistency, so S is maximally L-consistent. 
Theorem 3.6. If the set {B, A1 , A2 , ...} is L-consistent, so is {B, A1 , A2 , ...}.

Proof. We will proceed by contraposition. Assume that = {B, A1 , A2 , ...} is

inconsistent with L. This means that there is some finite whose elements imply a contradiction. Since any extension of an inconsistent set is itself
inconsistent, we can assume that contains B and reorder any Ai s in it so
= {B, A1 , A2 , ..., An }. From this we can generate the following proof:
Assumed
(B A1 A2 ... An ) =
(A1 A2 ... An ) = (B = )
((X Y ) = Z) is the same
as X = (Y = Z)
(A1 A2 ... An ) = B
(X = ) is equivalent to X
((A1 A2 ... An ) = B)
Necessitation
Schema K
(A1 A2 ... An ) = B
(A1 A2 ... An ) = B
Theorem 2.9
(A1 A2 ... An ) = (B = ) X = is equivalent to X
((X Y ) = Z) is the same
(B A1 A2 ... An ) =
as X = (Y = Z)
So {B, A1 , A2 , ..., } has an inconsistent finite subset and is therefore inconsistent itself. The theorem is thus proven by contraposition. 
Definition 3.7. (Canonical Model) We define the canonical model of L hW, R, i
as follows. Let W be the set of all maximally L-consistent sets of wffs. We define
the binary relation R on W as follows. Let , W . We say that R if and
only if for every proposition of the form X in , X is true in . Finally if P is a
propositional variable, P if and only if P .
Theorem 3.8. (Truth Lemma) Given a canonical model of L hW, R, i, a world
W , and a well-formed formula , the following is true:
if and only if .

Proof. Let be a world in the canonical model of L and let be a well-formed

formula. We will proceed by induction on the complexity of .
Atomic formulae. Say is a propositional variable P . By definition, P iff
P .
Negations. Say that our formula is , where is true in if and only if .
Say that . Then 6 , so 6 . Since is maximally consistent, it follows
that . Now say that . Since is consistent, 6 . By hypothesis it
follows that is not true in , so .
Modalities. Say that our formula is . First assume that  . Say is
accessible to . By the way we defined R we know that . Since our choice of

12

JOEL MCCANCE

was arbitrary, this holds for all the worlds to which is related. So  is true
in .
Now say that  6 . Since is maximally consistent, this means that  .
Consider the set of all statements beginning with , {X1 , X2 , ...}. We know
from the previous theorem that if {, X1 , X2 , ...} is consistent than so is
{, X1 , X2 , ...}. We can therefore extend this to form a maximally consistent set
. Now for every statement of the form P , P is in . So by our inductive
hypothesis, P . Since for every P P , we know that R by
definition. But since is true in , it must be that  is false in . So by
contraposition, if  is true in , then  .
Implications. Say that our formula is = and that and are each true if
and only if they are elements of . Say that = . Then either is false
in or is true. Say that is true, so must be true as well. Then and are
both in . Since they are both elements of the maximally a maximally consistent
set, = must be in that set as well.
Now assume that = . Then either or is in as well, since is
maximally consistent. Then either or , so = is true in .
Conjunctions. Say that our formula is and that and are true if and
only if they are elements of . Say that . Then both and must be
true, so , .
Now say that , . Then both and . So we know by definition
that . 
With this machinery in place, we can finally show that K is complete.
Theorem 3.9. Say that X is K-valid. Then there is a proof of X in the axiom
system K.
Proof. With the machinery above this proof becomes relatively straightforward.
We will proceed by the contrapositive. Assume that X has no proof in K. Since X
has no proof, the set {X} is consistent. (Since we cannot derive X, neither can
we derive X X.)
Extend this set to a maximally consistent set X and let hW, R, i be the canonical model of K. Clearly X , being maximally consistent, is in W . Since X X ,
X is true in X . Since X is maximally consistent, X 6 X . Therefore X is not
true in X , meaning that X is not valid in the canonical model.
Now clearly the canonical model of K is in the collection of frames K, since K
places no restrictions on its frames. We have therefore shown that there exists a
model based on a frame in K in which X is not valid. Therefore X is not K-valid.

To show that the remaining systems are true, we merely need to show that the
canonical model of a given system L meets the requirements of that system. For
example, the proof for T consists in showing that the canonical model of T is based
on a reflexive frame. However, it is helpful to begin with a quick lemma.
Lemma 3.10. Let hW, Ri be the canonical model for a system L and let W .
If contains no statements beginning with , then R for all W .
Proof. Fix W and let be the set of all statements beginning with  in .
Define = {X|X }. We know from the definition of the canonical model

A BRIEF INTRODUCTION TO MODAL LOGIC

13

that R only if . But since by hypothesis is empty, so must be . So

= . Ergo, is accessible to . 
Theorem 3.11. The systems D, T, K4, B, S4, and S5 are all complete.
Proof. As stated above, the only thing to do here is show that for any axiom system
L (where L is one of the above systems), the frame on which the canonical model
is based is in collection of frames L. In each of the following proofs let hW, R, i
be the canonical model of L, where L is the system under consideration.
(D) Fix W . Since we have taken D (P = P ) as an axiom of
D, we know that (P = P ) . Therefore P = P is true in
according to the Truth Lemma. Say that there was no W accessible to and
let X be a propositional variable. Since there are no worlds accessible to , X is
vacuously true. However, since D is true in , we know that X is in (since
is maximally consistent). Therefore X must be false in . But this can only
be the case if there exists accessible to such that 6 X. This contradicts
our earlier assumption, so it must be that there exists a accessible to . Since
this holds for all W , the canonical frame of D is serial.
(T) Fix W . Since we have taken T (P = P ) as an axiom of T, we
know that T is in . Therefore P = P is true in . If there are no statements
beginning with  in then R is true automatically. So say there are such
statements in and let = {X1 , X2 , ...} be the set of all statements beginning
with  in . Since axiom T is true in , each Xi is true in as well. So by the
definition of canonical models, R. Since our choice of was arbitrary, the frames
of T are reflexive.
(K4) Let , , W . Since we have taken axiom 4 (P = P ) as an
axiom of K4, we know that it must be an element of . So by the Truth Lemma,
P = P . Assume that R and R are true. If has no statements
beginning with  then we know from the lemma that R. So say there are such
statements in and let = {X1 , X2 , ...} be the set of all statements beginning
with  in . Since axiom 4 is true, we know that Xi is true in for each i N.
Fix such an i. As is accessible to , it follows that Xi . Since it must
therefore be true in and R, Xi must be an element of . So by the definition
of canonical models, must be accessible to . Therefore the canonical model of
K4 is transitive.
(B) The axiom system B includes axioms T and B (P = P ). We know
that T implies that hW, Ri is reflexive from above, so we simply need to show that
B guarantees symmetry. We will show this by contraposition. Let , W and
assume R. Say that X 6 . Since is maximally consistent, it follows that
X is in . Since we have assumed axiom B, we know then that X is true
in . Since is accessible to , this means that X is true in . But then we
know that X is not in . So by contraposition, if X is in then X is in . So
is accessible to by definition, making the canonical model of B symmetric.
(S4) The axiom system S4 includes axioms T and 4. Since we have already shown
that T and 4 guarantee reflexivity and transitivity, we know that the canonical
model of S4 is both reflexive and transitive.
(S5) The axiom system S5 includes axioms T , B, and 4. The above proofs
are therefore enough to show that S5 must therefore be reflexive, symmetric, and
transitive. 

14

JOEL MCCANCE

4. Quantified Modal Logic

While there are uses for propositional logic modal or otherwise we are as
mathematicians generally more interested in quantified or first-order logic. Without the quantification provided by and it is extremely difficult (perhaps impossible) to show general statements about such important mathematical structures as
the integers or the real numbers. However, adding quantifiers to modal logic turns
out to be a somewhat complex matter, even more so than adding them to PC. In
this last section we will illustrate the difficulties with an informal examination of
just what is required to determine the truth of a quantified modal statement.
The language for quantified modal logic is, as with the propositional case, an
expansion of the usual first-order logic. It consists of the following:
The same operators and punctuation as propositional modal logic;
The quantification symbols and ;
As many variables as we need, represented by lowercase Roman letters.
As many n-place relation symbols as we need, represented by capitalized
Roman letters.
A special relation =, the usual identity relation.

An atomic formula in this language will be any expression of the form R(x1 , x2 , .., xn ),
where R is an n-ary relation symbol, and build the rest of the formulas as you would
expect.
How might we construct a model for such a language? Since we are interested
in adding quantification, it would probably be useful to have some non-empty set
over which to quantify. We call this set the domain. We then determine the truth
of a formula like (x)(P (x)) in a world by checking to see if P (x) is true in for
every x in the domain. The evaluation for modalities also follows a similar route.
(x)(P (x)) can be determined true or false by checking to see if P (x) is true for
every value of x in every accessible world. Similarly, (x)(P (x)) says that there is
some element in the domain for which P (x) is true.
There is a danger on the horizon, however. Consider the veracity of (x)((x =
x)) in a world : there exists an x such that in every accessible world , x is
identical to itself. On the surface this seems quite readily true; how could it be that
something is not identical to itself? But there is a catch: how do we know whether
or not x actually exists in ? Does it even make sense to talk of objects that do
not exist? Yet we do not want to assert x 6= x, since that is a contradiction.
One possible solution is to simply require that if x is in one worlds domain, it
must be in them all. This is called a constant domain model, the frame of which
is a collection of worlds W , and accessibility relation R, and a single, non-empty
set D. However, it is not inherently unreasonable to want the domain to vary from
world to world. In temporal logic, for example, it would be ridiculous to say that
every object that exists now also existed 1,000 years ago. Temporal logic therefore
is better suited to a varying domain model. The frame of a varying domain model
looks identical to that of a constant domain model: a triple hW, R, Di. In this
case, however, D is a function assigning each world in W to a domain. The actual
domain for a world is not D, but D().
Unfortunately, this reintroduces the problem of deciding whether or not x = x
is true in if x 6 D(). One, somewhat unsatisfying option is to simply allow
that perhaps some statements are neither true nor false. This approach results in

A BRIEF INTRODUCTION TO MODAL LOGIC

15

what is called a partial model. Another solution is to define a frame domain

F , where F is just the union of all the domains. This allows us to talk about a
fixed x irrespective of whether or not it exists in the world in question. This way
we can define =So x is still identical to x whether or not it exists in a particular
world, since the relation = can just be defined as the set of all ordered pairs (y, y)
such that y F . This also solves the problem of whether P (x) is true or not in
when x 6 D(). Without the frame domain, P (x) would be vacuously true. With
the frame domain we can reasonably talk about (x) not being in the relation P , so
P (x) can be sensibly valued false.
Before concluding, let us consider how what is involved in showing the truth of
the statement (x)(P (x)) = (y)(P (y)) in a varying domain model. We have a
world , and its domain D(). We must examine all the elements of D() in order
to show the hypothesis of the statement. We must then fix our y and check P (y)
Conclusion
Modal logic provides logicians with a way to take into account not just the facts
themselves, but also how we come by these facts. While perhaps not useful in the
study of a priori mathematics, a rigorous treatment of modalities provides logicians
with the means to tease apart the implications of possibility, belief, and action on
the truth of judgments. However, a rigorous, symbolic treatment is required to open
up modal logic to the tools of mathematical logicians. By adding a single operator
to classical propositional logic, , we were able to develop such a treatment that
behaves as we would like a logical system to behave: only the true statements are
provable and all the provable statements are true. While the predicate case for
modal logic requires further development, the propositional case is both sound and
complete.
References
[1] Fitting, Melvin and Richard L. Mendelsohn, First-Order Modal Logic, The Netherlands:
Kluwer Academic Publishers, 1998.
[2] Hughes, G. E. and M. J. Cresswell, An Introduction to Modal Logic, London, Great Britain:
Spottiswoode, Ballantyne and Co Ltd: 1968.
[3] Hughes, G. E. and M. J. Cresswell, A New Introduction to Modal Logic, New York, NY:
Routledge, 1996.