05 pml

Lecture Notes on
Classical Modal Logic

15-816: Modal Logic
Andr´ Platzer
e

Lecture 5
January 26, 2010

1 Introduction to This Lecture
The goal of this lecture is to develop a starting point for classical modal
logic.
Classical logic studies formulas that are true (especially those that are
true in all interpretations, i.e., valid) and how truth is preserved in rea-
soning such that true premisses only have true consequences. These valid
formulas are characterized semantically as those that are true in all inter-
pretations I. We write I |= A if formula A holds in interpretation I. We
just write A if formula A holds in all interpretations and say that A is
valid. The most crucial criterion in classical logic is that logical reasoning
from valid assumptions should only lead us to valid conclusions for oth-
erwise there is something wrong with the reasoning schemes. Formulas of
the form A ∨ ¬A are always trivially valid in classical logic, because each
interpretation I satisfies either I |= A or I |= ¬A. Consequently, A ∨ ¬A.
Intuitionistic logic takes a more fine-grained view at logic and studies
formulas that are justified (by some argument) and how justification is pre-
served in reasoning. If there is a proof of A, intuitionistic logic would ac-
cept both A and A ∨ ¬A, but not without such a justification of either of the
two disjuncts. Thus, intuitionistic logic has a more fine-grained view than
just true/false signified in the classical axiom φ ∨ ¬φ or law of excluded
middle.
To some extent, modal logics also take a more fine-grained view. Clas-
sical model logics do not dispose off the law of excluded middle, though.

L ECTURE N OTES J ANUARY 26, 2010

L5.2 Classical Modal Logic

They still accept the axiom φ ∨ ¬φ and are a perfectly conservative exten-
sion of basic classical logic (they do not accept less formulas as valid than
classical propositional logic). But they allow distinctions between modes
of truth, i.e., between formulas that are true, necessarily true, possibly true,
possibly false, false. In fact, this similarity of intuitionistic logic and classi-
cal modal logic is not a mere coincidence, but can be made formally precise
by a translation of intuitionistic logic into classical modal logic where, ob-
viously, the new concept of necessity plays an important role.
The formal study of modal logic was founded by C. I. Lewis [Lew18].
Modal logic is an area with numerous results. As an excellent background
on modal logic, these notes are also partly based on a manuscript by Schmitt
[Sch03] and the book by Hughes and Cresswell [HC96]. Further back-
ground on modal logic can be found in the book by Fitting and Mendelsohn
[FM99]. Further material on the connections of modal an intuitionistic logic
can be found in [Fit83].

2 The Power of Knowledge in a Logic of Knowledge
Classical modal logics come in multifarious styles and variations. Here we
first introduce the classical propositional modal logic S4 and study varia-
tions later. We first follow an axiomatic approach to classical modal logic
and save the model-theoretic approach due to Kripke [Kri63] for later in
this course.
We start with an informal introduction and consider a well-known puz-
zle:

Three wise men are told to stand in a circle. A hat is put
in each of their heads. The hats are either red or black and ev-
eryone knows that there is at least one black hat. Every wise
man can see the color of the other hats except his own. They
are asked to deduce the color of their own hat without cheating
with a mirror or something of that sort. After some time went
by, one of the wise men says: “I don’t know which hat I have.”
With some more thought, another wise man says: “I don’t know
mine either.” “Then I know that my hat is black.” says the third
one.

The solution to this puzzle is a matter of knowledge, not just a matter of
truth. After the first wise man admits he doesn’t know the color of his hat,
the third wise man can conclude that wise men 2 and 3 cannot possibly


Classical Modal Logic L5.3

both have had red hats on, otherwise the ﬁrst white man would have seen
that and concluded that he must wear a black hat. The third wise man also
knows that the second wise man will be able to do the same reasoning and
know the same for he is wise. But once the second wise man admits he
doesn’t know the color of his hat either, the third wise man is now sure not
to wear a red hat.
Let us use the following propositional variables for i ∈ {1, 2, 3}:

Bi wise man i wears a black hat
Ri wise man i wears a red hat

We use a formula i φ to say that wise man number i knows that formula
φ holds true. Formula i φ clearly represents something else than φ being
true for φ might still be true, but wise man i may just not know that. The
operator i is what we call a modality.
What would we want to allow as valid reasoning schemes in a logic
of knowledge? Certainly, we want to allow all classical propositional rea-
soning, because our analysis is allowed to use all logical reasoning that we
know about in classical logic already. Modal instances of propositional tau-
tologies are perfectly acceptable. We want to accept i A → B ∨ i A and
i A ∨ ¬ i A, for instance. But what kind of reasoning with the knowledge
(or modalities) themselves do we admit?
The wise men know about all basic facts. This includes all tautologies
and all basic rules of the hat game, for instance, that there is at least one
black hat. We thus allow the proof rule called generalization rule:

φ
(G)
φ
We write this proof rule with modality in uni-modal logic. In the knowl-
edge logic case, we allow to use it for any instance where is replaced by
one of the i .
The wise men are truly wise and can draw conclusions. If a wise man
knows that he must wear a black hat if he doesn’t wear a red hat, and if he
knows that he does not wear a red hat, then he also knows that he wears a
black hat, because he is able to draw this conclusion. This is an instance of
the Kripke axiom:

(K) ( φ∧ (φ → ψ)) → ψ

If wise man i knows φ and he knows that φ implies ψ, then he also knows
consequence ψ, otherwise he wouldn’t be called a particularly wise man.



This axiom is often stated in the following elegant form, which is easily
obtained by propositional equivalences:

(K) (φ → ψ) → ( φ → ψ)

From these rules and axioms, we can easily derive the following rules
with G,K and modus ponens.

φ→ψ
(I)
φ→ ψ
φ↔ψ
(E)
φ↔ ψ

An extremely useful axiom that we can derive from the previous rules
and axioms is
( ∧) φ ∧ ψ ↔ (φ ∧ ψ)
The derivation is slightly more involved.
Now we know how to reason about knowledge, or at least have one
way of reasoning about who knows what, consider the wise men with hats.
The general facts from the puzzle are facts like B1 ∨ B2 ∨ B3 because there
is at least one black hat, and B1 → 2 B1 ∧ 3 B1 , because wise men 2 and
3 see and know if the first wise man wears a black hat. Moreover, the third
wise man knows that the second wise man can see the color of the first wise
guy and will know if it’s black: B1 → 3 2 B1 . Most importantly, the fact
that the first two wise men admit they do not know anything contributes
to what the third wise man will know. So we have:

¬ 1 B1 (1)
¬ 2 B2 (2)
¬ 1 R1
¬ 2 R2

From this the fact 3 B3 can be derived.
Although not necessary for solving the particular puzzle about the hats,
there are two more axioms that can make sense in a knowledge context:

(T) φ→φ

(4) φ→ φ



The first axiom T makes sense, because the wise men are wise: they
should only know things that are actually true. If the third wise man knows
that he has a black hat, then he should actually be wearing a black hat,
otherwise he is not very wise. So T could be called the “wise men only
know what’s true” axiom. When modeling belief rather than knowledge or
when modeling faulty knowledge, T will be dropped.
Axiom 4, instead, says that there is no passive knowledge. If a wise
man knows something, then he also knows that he knows it, and will not
say later on “oh I knew that but I just didn’t know I knew it”. Thus ax-
iom 4 represents an assumption on perfect and flawless knowledge and
introspection. The logic S4, for instance, is a classical modal logic with the
axioms K,T,4 and rules G, modus ponens and all propositional tautologies.
The meaning of set forth in this section is that of epistemic modal
logic in a logic of knowledge. Formula φ is taken to mean that some
entity “knows φ”.

3 Classical Propositional Modal Logic
Let Σ be a set of propositional letters or atomic propositions. The syntax of
classical propositional modal logic is defined as follows:

Definition 1 (Propositional modal formulas) The set FmlPML (Σ) of formu-
las of classical propositional modal logic is the smallest set with:

• If A ∈ Σ is a propositional letter, then A ∈ FmlPML (Σ).

• If φ, ψ ∈ FmlPML (Σ), then ¬φ, (φ ∧ ψ), (φ ∨ ψ), (φ → ψ) ∈ FmlPML (Σ).

• If φ ∈ FmlPML (Σ) and x ∈ V , then ( φ), (♦φ) ∈ FmlPML (Σ).

The informal meaning of φ would be that φ is necessary (holds in
all possible worlds). Formula ♦φ, instead, would mean that φ is possible
(holds in some possible world). This is the alethic meaning of , where φ
is taken to mean that φ is necessary.
For reference, Figure 1 summarizes the axioms and rules we have iden-
tified for modal logic so far: The Kripke axiom K, the T axiom, the 4 axiom,
¨
modus (ponendo) ponens MP, and the Godel or necessitation rule G. Note,
however, that there are many different variations of modal logic.
We say that a formula ψ is provable or derivable from a set of formulas
if there is a Hilbert-style proof:



(P) all propositional tautologies

(K) (φ → ψ) → ( φ → ψ)

(T) φ→φ

(4) φ→ φ

φ φ→ψ
(MP)
ψ
φ
(G)
φ

Figure 1: Modal logic S4

Definition 2 (Provability) Let S be a system of modal logic, i.e., a set of proof
rules (including axioms) like, e.g., S4. For a formula ψ and a set of formulas Φ,
we write Φ S ψ and say that ψ can be derived from Φ (or is provable from Φ), iff
there is a proof of ψ that uses only the formulas of Φ and the axioms and proof rules
of S. That is, we define Φ S ψ inductively as:

Φ S ψ

iff ψ ∈ Φ or there is an instance

φ1 ... φn
ψ

of a proof rule of S with conclusion ψ and some number n ≥ 0 of premisses such
that for all i = 1, . . . , n, the premiss φi is derivable, i.e.:

Φ S φi

Note that the case n = 0 is permitted, which corresponds to axioms.

4 Godel Translation
¨
Intuitionistic logic takes a more fine-grained view than classical truth or
false with its law of excluded middle or tertium-non-datur. In classical
(two-valued) logic, where the central constructions are about truth and



preservation of truth, every formula is either true or false in a given in-
terpretation. In particular, A ∨ ¬A is a classical tautology for A either has
to be true or false.
In intuitionistic logic, the central constructions are about justification
and preservation of justification. For the formula A ∨ ¬A, there is (usually)
no justification of A, nor a justification of ¬A. The law of excluded middle
is thus not accepted.
In the realm of modal logics, however, there is a way to understand
intuitionistic logic in a modal setting. After all, modal logic also takes a
more fine-grained view of modes of truth.
The intuition behind understanding intuitionistic logic in a classical set-
ting is to identify intuitionistic truth (being justified) with classical prov-
ability. The Godel translation G maps formulas of intuitionistic logic to
¨
modal logic by prefixing all formulas with the modality , which is under-
stood as “provable”. This translation G is defined inductively:

G(a) = a if a ∈ Σ is a propositional letter
G(φ ⊃ ψ) = G(φ) → G(ψ)
G(φ ∧ ψ) = G(φ) ∧ G(ψ)
G(φ ∨ ψ) = G(φ) ∨ G(ψ)

Translation G captures the idea that we would accept a in an intuitionistic
setting if a is provable. Likewise, we would accept an intuitionistic impli-
cation φ ⊃ ψ if the (translated) implication G(φ) → G(ψ) is provable.
The question is, if there is a way to characterize the formulas obtained
by Godel translation G from provable formulas of intuitionistic logic. In
¨
fact, it turns out that an intuitionistic propositional formula is provable
(intuitionistically) if and only if its translation is provable in propositional
modal logic, provided that we have the right set of axioms. What proper-
ties should satisfy for a provability interpretation?
¨
With the provability interpretation for the Godel translation, we expect
that the K axiom makes sense. If φ → ψ is provable, and φ is provable, then
we should be able to glue their proofs together to a proof of ψ:

(K) (φ → ψ) → ( φ → ψ)

Moreover, we expect to be able to prove only properties that are actually
true, otherwise we would not venture to call it a proof. Thus if φ is prov-
able, it should be true:
(T) φ→φ



If a formula is provable, then it should be provable that it is provable, for
the proof itself already is a very good proof of provability. If φ is provable,
then it should be provably provable:

(4) φ→ φ

Provability is a rational notion, so we expect the notion to be closed both
under arbitrary propositional inferences and the modus ponens. After all,
these only glue together proofs:

φ φ→ψ
(MP)
ψ

Finally, if we have proven any formula φ, then it should be provable, for
otherwise, we would not call it proven:

φ
(G)
φ

In summary, the axioms and rules we need in this provability interpre-
tation of directly coincide with those of the modal logic S4, i.e., Figure 1.
In fact, it can be shown that an intuitionistic formula F is provable in
intuitionistic logic if and only if their translation GF is provable in S4. The
proof of this statement requires more techniques than we have at this stage
of the lectures.

5 Kripke Structures
Another introduction to modal logic follows transition systems and ﬁnite
automata.
Consider the example of a transition structure in Figure 2. The names of
the state are not of relevance to us here, only what values two signals or in-
ternal state variables have in these states. We consider those state variables
as propositional variables p and q. Their actual values in the respective
states of the transition system are as indicated in Figure 2. For this tran-
sition system, we want to express that p is false in all successor states of a
state in which both p and q are true. Likewise, p is still false in all successors
of all successors of states in which p and q are true. This property does not
generalize to all third successor states though. Similarly, if p and q are both
false, then p is true in all successor states.



truth-value of p
TT TF truth-value of q

pq
FT FF

Figure 2: A transition system

In order to formalize these properties, propositional logic is not quite
accurate, because it is not only important what is true and false, but also in
which states something is true and false. In addition, the notion of succes-
sor states or a means to refer to them does not exist in propositional logic.
Now consider the modality with the intended semantics being that φ
holds true in a state, if φ holds true in all of its successors. The modality ♦
would be taken to mean that ♦φ holds true in a state, if φ holds true in at
least one of its successors. Then we can phrase the above properties quite
naturally:

p∧q → ¬p
p∧q → ¬p
p∧q → ¬p
¬p ∧ ¬q → p
¬p ∧ ¬q → p
¬p ∧ ¬q → p

Note that the nesting of refers to all successors of all successors (double
nesting), or all successors of all successors of all successors (triple nesting),
respectively. Some of these formulas are true in some states of Figure 2, oth-
ers are true in all states of Figure 2. Yet another class of formulas may even
be true in all states of all transition systems, and not just in the particular
transition system depicted in Figure 2.



6 Kripke Semantics
The meaning of formulas in propositional modal logic is defined in terms
of truth in possible worlds, due to Kripke [Kri63], following suggestions of
Leibniz for the understanding of necessity as truth in all possible worlds.
An interpretation consists of a non-empty set W of possible worlds. For
each world s ∈ W we need an assignment of a truth-value to each propo-
sitional letter A ∈ Σ. The notions of possibility and necessity depend on
which worlds are possible or conceivable from which other world. For
that, an interpretation also consists of an accessibility relation ρ ⊆ W × W
among worlds. The relation (s, t) ∈ ρ would hold if world t is accessible
from world s. Interchangeably, we also write just sρt iff (s, t) ∈ ρ. A dif-
ferent way to explain ρ is that it defines—from the perspective of world
s—which world t is possible or conceivable.

Definition 3 (Kripke frame) A Kripke frame (W, ρ) consists of a non-empty
set W and a relation ρ ⊆ W × W on worlds. The elements of W are called possible
worlds and ρ is called accessibility relation.

Definition 4 (Kripke structure) A Kripke structure K = (W, ρ, v) consists
of Kripke frame (W, ρ) and a mapping v : W → Σ → {true, false} that assigns
truth-values to all the propositional letters in all worlds.

By an abuse of notation, you will sometimes find the notation s(A) instead
of v(s)(A). See exercise.

Definition 5 (Interpretation of propositional modal formulas) Given a Kripke
structure K = (W, ρ, v), the interpretation |= of modal formulas in a world s is
defined as

1. K, s |= A iff v(s)(A) = true.

2. K, s |= φ ∧ ψ iff K, s |= φ and K, s |= ψ.

3. K, s |= φ ∨ ψ iff K, s |= φ or K, s |= ψ.

4. K, s |= ¬φ iff it is not the case that K, s |= φ.

5. K, s |= φ iff K, t |= φ for all worlds t with sρt.

6. K, s |= ♦φ iff K, t |= φ for some world t with sρt.

When K is clear from the context, we also often abbreviate K, s |= φ by K, s |= φ.



Definition 6 (Validity) Given a Kripke structure K = (W, ρ, v), formula φ is
valid in K, written K |= φ, iff K, s |= φ for all worlds s ∈ W .

Let K be the Kripke structure corresponding to Figure 2, then

K |= p ∧ q → ¬p
K |= p ∧ q → ¬p
K |= p ∧ q → ¬p
K |= ¬p ∧ ¬q → p
K |= ¬p ∧ ¬q → p
K |= ¬p ∧ ¬q → p
K |= ¬p ∧ q → ♦p
K |= ¬p ∧ q → ♦¬p
K |= ¬p ∧ q → ♦(¬p ∧ q)
K |= ¬(p ↔ q) → ♦¬(p ↔ q)
K |= ¬(p ↔ q) → ¬ ¬(p ↔ q)
K |= (p ↔ q) → ¬♦(p ↔ q)

7 Consequences
For defining consequences of formulas in modal logic, we need to distin-
guish if the assumptions are meant to hold locally in the current world, or
globally for all worlds.
Definition 7 (Local consequence) Let ψ be a formula and Φ a set of formulas.
Then we write Φ l ψ if and only if, for each Kripke structure K = (W, ρ, v) and
each world s ∈ W :
K, s |= Φ implies K, s |= ψ
Likewise, we write Φ C ψ if the local consequence holds for all Kripke
l
structures of a class C (instead of all Kripke structures by and large). This
will be of relevance if we are not interested in all Kripke structures but only
those of a certain shape, say, all reflexive Kripke structures.
Definition 8 (Global consequence) Let ψ be a formula and Φ a set of formulas.
Then we write Φ g ψ if and only if, for each Kripke structure K = (W, ρ, v):

if for all world s ∈ W : K, s |= Φ



then
for all world s ∈ W : K, s |= ψ
Again, we write Φ C ψ if the global consequence holds for all Kripke struc-
g
tures of a class C.

Definition 9 (Tautology) A formula φ is valid or a tautology, iff ∅ l φ, which
we write φ. A set of formulas Φ is called satisfiable, iff there is a Kripke struc-
ture K and a world s with K, s |= Φ.

Again, we write C φ if formula φ is valid for all Kripke structures of a class
C.

Lemma 10 (Local deduction theorem) For formulas φ, ψ we have

φ l ψ iff l φ→ψ

8 Modal Logic and Finite Automata
Consider the finite automaton in Figure 3 over the alphabet {0, 1} with ini-
tial state p and accepting state F . Consider its corresponding transition

start 1
0
p q 1 1
1 F s 0,1
1
0,1

Figure 3: A finite automaton / acceptor

structure as a Kripke structure, where the assignment of propositional let-
ter at states is as indicated. That is, at the left-most state only propositional
letter p holds, at the right-most, only s holds and so on. With this, the states
of the finite automaton are captured in the Kripke structure.
The finite automaton has labels on the edges also, which cannot (really)
be captured in the states. Instead, we consider a labelled transition struc-
ture where the input 0,1 is represented as labels on the accessibility relation.
Now we have two accessibility relations ρ(0) and ρ(1) for the accessibility
under input 0 and under input 1, respectively. To access these two sepa-
rate accessibility relations in logical formulas, we use two separate pairs of



modalities, which are also labelled with input 0 or input 1, respectively: the
modality pair 0 and ♦0 referring to the accessibility relation ρ(0), and the
modality pair 1 and ♦1 for the accessibility relation ρ(1).
Let K be the Kripke structure corresponding to Figure 3, then

K |= ¬♦0 F does not end with 0
K |= p → ♦0 p p has a 1-loop
K |= ♦0 true never stuck with input 0
K |= ♦1 true never stuck with input 1
K |= F → 0 (¬♦0 F ∧ ¬♦1 F ) no end one step after seeing 0 from F

The last formula is a bit cumbersome to write. So we introduce a third pair
of modal operators 01 and ♦01 that we bind to refer to transition under
any input (0 or 1) by assuming the following axiom (for all instantiations
of formula φ):

♦01 φ ↔ ♦0 φ ∨ ♦1 φ

With this we ﬁnd that:

K |= F → 0 ¬♦01 F no end one step after seeing 0 from F
K |= F → 0 ¬♦01 ♦01 F no end two steps after seeing 0 from F
K |= p → ♦01 q p has a q successor
K |= F → 1F stay ﬁnal on 1s

Supposing we do not know the transition system, but only the above
modal formulas. What other formulas can we infer about the system? Let
us assume the following set of formulas Γ:

¬♦0 F
p → ♦0 p
♦0 true
♦1 true
F → 0 (¬♦0 F ∧ ¬♦1 F )



Can we conclude any of the following consequences?
?
Γ l F → ♦1 F ?
?
Γ g F → ♦1 F ?
?
Γ l F → ♦1 ♦1 F ?
?
Γ g F → ♦1 ♦1 F ?

It turns out that the first two consequences hold using F → 1 F and ♦1 true
from Γ. The third one is not a consequence, because the local facts are
not sufficient. The fourth consequence, instead, is justified using again
F → 1 F and ♦1 true from Γ, but needs these facts globally.
Another question is if we can characterize the finite automaton in Fig-
ure 3 using a finite set of modal formulas?



Exercises
Exercise 1 Give a Hilbert-proof for the property 3 B3 from the facts and rules
in Section 2. Please prove this much(!) more systematically than in the informal
introduction in class. Is there a contradiction because the first wise man would be
able to conclude a fact like the third one did, after the second wise man announced
¬ 2 B2 ? Discuss how knowledge would change if the task for the wise guys would
be to deduce the answer of any wise man, rather than the color of their own hats.
Would this different setting still make sense?
Exercise 2 In the definition of Kripke structures you will sometimes find that v
is not mentioned and that the notation s(A) is used instead of v(s)(A). Hence,
the truth-value of the propositional variables is associated with the state. Does this
make a difference? If so, give an example where the difference can be seen and
explain why. If not, prove that the the original and the new semantics are actually
equivalent.
Exercise 3 Prove or disprove that the following formulas are modal tautologies.
If you disprove it, also try to find a variation of the formula or a class of Kripke
structures for which you can prove it.
1. φ∧ (φ → ψ) → ψ
2. φ → ♦φ
3. φ → φ
4. φ ↔ ¬♦¬φ
5. (φ ∧ ψ) ↔ ( φ ∧ ψ)
6. (φ ∨ ψ) ↔ ( φ ∨ ψ)
7. ♦(φ ∧ ψ) ↔ (♦φ ∧ ♦ψ)
8. ♦(φ ∨ ψ) ↔ (♦φ ∨ ♦ψ)
9. φ → ♦φ
Exercise 4 How does the following variation H of the G¨ del translation affect the
o
results
H(a) = a if a ∈ Σ is a propositional letter
H(φ ∧ ψ) = H(φ) ∧ H(ψ)
H(φ ∨ ψ) = H(φ) ∨ H(ψ)
H(φ ⊃ ψ) = H(φ) → H(ψ)



Do G and H share the same properties or is there an important difference? Does it
establish a different connection to intuitionistic logic or the same? Do we need the
same axioms and rules or not? Prove or disprove each of these conjectures.

Exercise 5 Prove or disprove both directions of the local deduction theorem Lemma
10.

Exercise 6 Prove or disprove both directions of the variation of deduction theorem
Lemma 10 with l replaced by g .



References
[Fit83] Melvin Fitting. Proof Methods for Modal and Intuitionistic Logic. Rei-
del, 1983.

[FM99] Melvin Fitting and Richard L. Mendelsohn. First-Order Modal
Logic. Kluwer, Norwell, MA, USA, 1999.

[HC96] G.E. Hughes and M.J. Cresswell. A New Introduction to Modal
Logic. Routledge, 1996.

[Kri63] Saul A. Kripke. Semantical considerations on modal logic. Acta
Philosophica Fennica, 16:83–94, 1963.

[Lew18] Clarence Irving Lewis. A Survey of Symbolic Logic. University of
California Press, Berkeley, 1918. Republished by Dover, 1960.

[Sch03] Peter H. Schmitt. Nichtklassische Logiken. Vorlesungsskriptum
a ¨
Fakult¨ t fur Informatik , Universit¨ t Karlsruhe, 2003.
a


05 pml

More Related Content

What's hot (19)

Viewers also liked (11)

Similar to 05 pml (20)

Recently uploaded (20)

05 pml