An Introduction to Modal Logic

2009 Formosan Summer School on Logic, Language, and Computation

29 June-10 July, 2009

FLOLAC2009@NTU 1
 The Agenda
• Introduction
• Basic Modal Logic
• Normal Systems of Modal Logic
• Meta-theorems of Normal Systems
• Variants of Modal Logic
• Conclusion

FLOLAC2009@NTU 2
 Introduction
Let me tell you the story

FLOLAC2009@NTU 3
 Introduction
• Historical overview
• Conceptual overview
• Further readings

FLOLAC2009@NTU 4
 Historical Overview
• Pre-history
• The syntactic era (1918-1959)
• The classical era (1959-1972)
• The modern era (1972-present)

FLOLAC2009@NTU 5
 Pre-history: How It Starts

The Battle of Salamis, 20 September, 480 B.C

FLOLAC2009@NTU 6
 Pre-history: Aristotle’s De Interpretatione 9
• the problem of future contingents: a logical paradox by
Diodorus Cronus, Megarian school of philosophy
• what happens was necessarily going to happen
• what does not happen was necessarily going to not happen
• Aristotle: statements about the future are neither true nor false

FLOLAC2009@NTU 7
 Pre-history: Aristotle’s De Interpretatione 9
. . . if a thing is white now, it was true before to say that it
would be white, so that of anything that has taken place,
it was always true to say ‘it is’ or ‘it will be’. But if it
was always true to say that a thing is or will be, it is not
possible that it should not be or not come to be, and when a
thing cannot not come to be, it is impossible that it should
not come to be, and when it is impossible that it should not
come to be, it must come to be. All then, that is about to
be must of necessity take place. It results from this that
nothing is uncertain or fortuitous, for if it were fortuitous it
would not be necessary.
[Translation, Ross, 1928]

FLOLAC2009@NTU 8
 Pre-history: Aristotle’s De Interpretatione 9
1. ϕ: there is (was) a sea-battle on 20 September, 480 B.C.
2. ϕ ⊃ ϕ
3. ¬ϕ ⊃ ¬ϕ
4. ϕ ∨ ¬ϕ
5. ϕ ∨ ¬ϕ
6. logical validity is universal, so ϕ ∨ ¬ϕ holds before 20
September, 480 B.C: fatalism

FLOLAC2009@NTU 9
 Pre-history: Aristotle’s Square of Opposition

A : p, E : ¬p, I : ♦p, O : ♦¬p

contrary: ¬(p ∧ ¬p) subaltern: p ⊃ ♦p, ¬p ⊃ ♦¬p

subcontrary: ♦p ∨ ♦¬p contradictory: p ≡ ¬♦p, ¬p ≡ ¬♦p

FLOLAC2009@NTU 10
 Pre-history: Possible Worlds
• G.W. Leibniz (1646/7/1-1716/11/14):
– a possible world is made up of individuals that are compos-
sible – that is, individuals that can exist together.
– possible worlds exist as possibilities in the mind of God.
– one world among them is realized as the actual world, and
this is the most perfect one

FLOLAC2009@NTU 11
 Pre-history: Possible Worlds
• modal status of a proposition:
– truth: true in the actual world
– falsity: false in the actual world
– possibility: true in at least one possible world
– impossibility: true in no possible world
– necessity: true in all possible worlds
– contingency: true in some possible worlds and false in others

FLOLAC2009@NTU 12
 Pre-history

Aristotle and G.W. Leibniz

FLOLAC2009@NTU 13
 Historical Overview
• Pre-history
• The syntactic era (1918-1959)
• The classical era (1959-1972)
• The modern era (1972-present)

FLOLAC2009@NTU 14
 The Syntactic Era
• paradox of material implication (⊃):
1. ¬ϕ ` (ϕ ⊃ ψ)
2. ψ ` (ϕ ⊃ ψ)
3. (ϕ ∧ ψ) ⊃ χ ` (ϕ ⊃ χ) ∨ (ψ ⊃ χ)
4. (ϕ1 ⊃ ϕ2) ∧ (ψ1 ⊃ ψ2) ` (ϕ1 ⊃ ψ2) ∨ (ψ1 ⊃ ϕ2)
5. ϕ ⊃ ψ ` ϕ ∧ χ ⊃ ψ
6. ϕ ⊃ ψ, ψ ⊃ χ ` ϕ ⊃ χ
7. ϕ ⊃ ψ ` ¬ψ ⊃ ¬ϕ
8. ¬(ϕ ⊃ ψ) ` ϕ
9. ` (ϕ ⊃ ψ) ∨ (¬ϕ ⊃ ψ)
10. ` (ϕ ⊃ ψ) ∨ (ψ ⊃ ϕ)
11. ¬ϕ ⊃ ¬(ψ ⊃ χ), ¬ψ ` ϕ
FLOLAC2009@NTU 15
 The Syntactic Era
• there are counterintuitive results to formulate “if ϕ, then ψ”
as ϕ ⊃ ψ
• Dorothy Edgington’s proof of the existence of God
– ϕ: God exists; ψ: I pray; χ: my prayers will be answered
– ¬ϕ ⊃ ¬(ψ ⊃ χ), ¬ψ ` ϕ
– if God does not exist, then it’s not the case that if I pray, my
prayers will be answered; and I don’t pray; so God exists!
• strict Implication (C.I. Lewis, 1912): ‘it is necessarily the case
that ϕ implies ψ
ϕ → ψ =def (ϕ ⊃ ψ)

FLOLAC2009@NTU 16
 The Syntactic Era
• axiomatic systems: S1 to S5 by Lewis
• proving distinctness theorems
• lack of natural semantics
• three lines of work to next stage:
– Carnap’s state description (close to possible world seman-
tics)
– Prior’s tense logic: with semantic ideas and insights (the
model (ω, <))
– Jónsson and Tarski: representation theorem of modal alge-
bra (an algebraic analog of the canonical model techniques)

FLOLAC2009@NTU 17
 The Syntactic Era

C.I. Lewis, R. Carnap, A.N. Prior, and A. Tarski

FLOLAC2009@NTU 18
 Historical Overview
• Pre-history
• The syntactic era (1918-1959)
• The classical era (1959-1972)
• The modern era (1972-present)

FLOLAC2009@NTU 19
 The Classic Era
• Kripke semantics: accessibility relation (Kripke, 1959; Hintikka,
1957; Kanger, 1957)
• canonical model, completeness, filtration (Lemmon and Scott,
1977)
• relational structures: as analytic tools, not really to be described
• many applications in modeling of agents

FLOLAC2009@NTU 20
 The Classic Era

S.A. Kripke, J. Hintikka, S. Kanger, and D. Scott

FLOLAC2009@NTU 21
 Historical Overview
• Pre-history
• The syntactic era (1918-1959)
• The classical era (1959-1972)
• The modern era (1972-present)

FLOLAC2009@NTU 22
 The Modern Era
• frame incompleteness (Thomason, 1972, 1974)
• Sahlqvist correspondence theorem (1973)
• universal algebra: algebraic semantics
• classical model theory: correspondence theory, bisimulation
(van Benthem)
• connections with other fields (Gabbay, Halpern, et al.):
– computer science and AI: dynamic logic, description logic,
temporal logic, epistemic logic, complexity
– economics: game logic, (dynamic) epistemic logic
– mathematics: co-algebra, non-well-founded set theory, ge-
ometry, topology
– linguistics: feature logic
FLOLAC2009@NTU 23
 The Modern Era

J. van Benthem, D. Gabbay, and J. Halpern

FLOLAC2009@NTU 24
 Introduction
• Historical overview
• Conceptual overview
• Further readings

FLOLAC2009@NTU 25
 Philosophical Background
• the monotheistic approach: choosing one of all possible logical
languages and saying ‘This is THE Logic’
• the polytheistic approach: as a discipline that investigate dif-
ferent logical languages.

FLOLAC2009@NTU 26
 The Polytheistic Approach to Modal Logics
• alethic modal logic: necessity and possibility
• epistemic/doxastic logic: knowledge/belief
• deontic logic: obligation, permission, prohibition
• dynamic logic: action, program
• temporal logic: tense (future, past, since, until)
• description logic: role (universal and existent), Web 3.0, ontol-
ogy language
• arrow logic, spatial logic, µ-calculus, game logic, coalition logic,
dynamic epistemic logic etc.

FLOLAC2009@NTU 27
 Three Slogans
• modal languages are simple yet expressive languages for talking
about relational structures
• modal languages provide an internal, local perspective on rela-
tional structures
• modal languages are not isolated formal systems
– languages: modal vs classical (FOL,SOL), internal vs exter-
nal perspective
– relational structures vs Boolean algebra with operators
(BAO): Jónsson and Tarski’s representation theorem
(From P. Blackburn, M. De Rijke, and Y. Venema, Modal Logic)

FLOLAC2009@NTU 28
 The Slogans in Reality
• example: trust and reputation management in P2P systems
• internal and local perspective
– each individual (peer) looks at itself and its neighbor
– each neighbor is inside the community
– distributed management by each individual
• external and global perspective
– a central authority looks at the whole community
– the authority is outside the community
– centralized management by the authority

FLOLAC2009@NTU 29
 Introduction
• Historical overview
• Conceptual overview
• Further readings

FLOLAC2009@NTU 30
 Standard Textbooks and References
• B.F. Chellas. Modal Logic: An Introduction. Cambridge University Press, 1980.
• G. Hughes and M.J. Cresswell. A Companion to Modal Logic. Methuen, 1984.
• G. Hughes and M.J. Cresswell. A New Introduction to Modal Logic. Routledge, 1996.
• A. Chagrov and M. Zakharyaschev. Modal logic. Oxford University Press, 1997.
• P. Blackburn, M. De Rijke, and Y. Venema. Modal Logic. Cambridge University Press,
2001.
• P. Blackburn, J. van Benthem, and F. Wolter. Handbook of Modal Logic. Elsevier,
2007.

FLOLAC2009@NTU 31
 Internet Resources
• Advances in Modal Logic
• Stanford Encyclopedia of Philosophy
• Preview on the Handbook of Modal Logic
• Logic and Rational Interaction
• List of Resources in Wikipedia

FLOLAC2009@NTU 32
 Basic Modal Logic
Let me show you the diamond(s)

FLOLAC2009@NTU 33
 Basic Modal Logic
• Basic Modal Logic—Syntax
• Basic Modal Logic—Semantics
• Basic Model Theory

FLOLAC2009@NTU 34
 Basic Modal Logic—Syntax
• alphabet
– a set of atomic propositional variables Φ0 = {p1, p2, · · ·}
– primitive logical symbols: ⊥ (contradiction), ¬ (negation),
∧ (conjunction), ♦ (possibility modality)
– defined logical symbols: > (tautology), ∨ (disjunction),
⊃ (material implication), ≡ (equivalence),  (necessity
modality)
– auxiliary symbols: (, )
• well-formed formulas (wff):
– any atomic propositional variable is a wff
– ⊥ is a wff
– if ϕ and ψ are wffs, so are :¬ϕ, ♦ϕ, and ϕ ∧ ψ
FLOLAC2009@NTU 35
 Basic Modal Logic—Syntax
• abbreviations:
1. >: ¬⊥
2. ϕ ∨ ψ: ¬(¬ϕ ∧ ¬ψ)
3. ϕ ⊃ ψ: ¬ϕ ∨ ψ
4. ϕ ≡ ψ: (ϕ ⊃ ψ) ∧ (ψ ⊃ ϕ)
5. ϕ: ¬♦¬ϕ

FLOLAC2009@NTU 36
 Several Modalities

John successful.
• is necessarily
• is possibly
• is believed/known to be
• is permitted to be
• ought to be
• is now
• will be
• has a strategy to become
• ······
FLOLAC2009@NTU 37
 Conventions of Modalities

Modalities Alethic epistemic doxastic deontic temporal dynamic description

logic logic1 logic2 logic3 logic4 logic5 logic
Possibility ♦ ¬K¬ ¬B¬ P F, P hαi ∃R
Necessity  K B O G, H [α] ∀R
Impossibility ¬♦ K¬ B¬ F

1. K: know
2. B: believe
3. P : permission, O: obligation, F : prohibition (forbid)
4. F (P ): future (past) possibility, G(H): future (past) necessity
5. [α](hαi): it’s necessary (possible) after the execution of α

FLOLAC2009@NTU 38
 Syntax and Semantics
• Basic Modal Logic—Syntax
• Basic Modal Logic—Semantics
• Basic Model Theory

FLOLAC2009@NTU 39
 Basic Modal Logic—Semantics
• (Kripke) frame: F = (W, R)
– W : a set of possible worlds (points, states, etc.)
– R ⊆ W × W : a binary relation over W
• (Kripke) model: M = (F, π) = (W, R, π)
– F: a frame
– π : Φ0 → 2W : a truth assignment
• intuition
– a frame is a (very basic) relational structure
– π(p): the set of all worlds in which p is true
– w ∈ π(p): the atomic proposition p is true in the world w
– w 6∈ π(p): the atomic proposition p is false in the world w
FLOLAC2009@NTU 40
 Basic Modal Logic—Semantics
• satisfaction: given a model M = (W, R, π) and w ∈ W
1. M, w 6 ⊥
2. M, w p ⇔ w ∈ π(p) for any p ∈ Φ0
3. M, w ¬ϕ ⇔ M, w 6 ϕ
4. M, w ϕ ∧ ψ ⇔ M, w ϕ and M, w ψ
5. M, w ♦ϕ ⇔ there exists u ∈ W such that (w, u) ∈ R
and M, u ϕ

FLOLAC2009@NTU 41
 Basic Modal Logic—Semantics
• derived rules of satisfaction
1. M, w >
2. M, w ϕ ∨ ψ ⇔ M, w ϕ or M, w ψ
3. M, w ϕ ⊃ ψ ⇔ if M, w ϕ then M, w ψ
4. M, w ϕ ≡ ψ ⇔ M, w ϕ iff M, w ψ
5. M, w ϕ ⇔ for all u ∈ W such that (w, u) ∈ R,
M, u ϕ

FLOLAC2009@NTU 42
 Basic Modal Logic—Semantics
• M, w ϕ: ϕ is satisfied or true in M at state w
• M, w 6 ϕ: ϕ is refuted or false in M at w
• M ϕ: ϕ is globally or universally true in M: for all w ∈ W ,
M, w ϕ
• ϕ is satisfiable in M if there exists w ∈ W s.t. M, w ϕ
• ϕ is satisfiable in a class of models if it is satisfiable in some
model belonging to the class
• ϕ is falsifiable or refutable if ¬ϕ is satisfiable

FLOLAC2009@NTU 43
 Basic Modal Logic—Semantics
• let Σ denote a set of wffs
• M, w Σ: for all ϕ ∈ Σ, M, w ϕ
• M Σ: Σ is globally or universally true in M: for all w ∈ W ,
M, w Σ
• Σ is satisfiable in M if there exists w ∈ W s.t. M, w Σ

FLOLAC2009@NTU 44
 Basic Modal Logic—Semantics
• C: a class of Kripke models, Σ: a set of wffs, ϕ: a wff
• Σ C ϕ:
– ϕ is a local semantic consequence of Σ over C
– for all models M in C and all worlds w in M, if M, w Σ,
then M, w ϕ
• Σ gC ϕ:
– ϕ is a global semantic consequence of Σ over C
– for all models M in C, if M Σ, then M ϕ
• C ϕ: ∅ C ϕ: ϕ is valid on C
• note: {p} 6 C p, but {p} gC p

FLOLAC2009@NTU 45
 Properties of Binary Relations and Models
• a binary relation R is
– serial: ∀w∃u, (w, u) ∈ R
– reflexive: ∀w, (w, w) ∈ R
– symmetric: ∀w, u, (w, u) ∈ R ⇒ (u, w) ∈ R
– transitive:
∀w, u, v, (w, u) ∈ R ∧ (u, v) ∈ R ⇒ (w, v) ∈ R

– Euclidean:
∀w, u, v, (w, u) ∈ R ∧ (w, v) ∈ R ⇒ (u, v) ∈ R

• a frame F = (W, R) or a models M = (W, R, π) is serial (resp.

reflexive, symmetric, transitive, Euclidean) if R is
FLOLAC2009@NTU 46
 Example I: A Kripke Frame

F = (W, R)
• W = {w0, w1, w2, w3}
• R = {(w0, w0), (w0, w3), (w3, w0), (w3, w3),
(w1, w0), (w1, w3), (w2, w0), (w2, w3)}
FLOLAC2009@NTU 47
 Example I: A Kripke Frame

the frame may be

• a graph
• a social network
• a communication network
• a state transition system
• ······
FLOLAC2009@NTU 48
 Example I: A Scenario Based on the Frame
• the frame is a state transition system
• the scenario
– there are two political parties —BLUE and GREEN— in a
country
– the country may be in two situations — BAD and GOOD
• atomic propositions
– blue: the ruling party of the country is BLUE
– green: the ruling party of the country is GREEN
– bad: the country is in a BAD situation
– good: the country is in a GOOD situation

FLOLAC2009@NTU 49
 Example I: A Kripke Model

• (blue, green, bad, good) • π(green) = {w1, w3}

• M = (W, R, π) • π(bad) = {w0, w1}
• π(blue) = {w0, w2} • π(good) = {w2, w3}
FLOLAC2009@NTU 50
 Example I: A Kripke Model

R = {(w, w0), (w, w3) | w ∈ W }

• M, w0 bad • M, w0 green ⊃ good
• M, w0 ¬green • M, w3 green ⊃ good
• M, w3 good • M, w (blue ⊃ bad)
• M, w3 ¬blue • M, w (green ⊃ good)
• M, w0 blue ⊃ bad • M (blue ⊃ bad)
• M, w3 blue ⊃ bad • M (green ⊃ good)

∀π, ϕ, (F, π) ϕ ⊃ ϕ

FLOLAC2009@NTU 51
 Example II: Logical and Physical Possibility
• physical constraint: the speed limit of T700 train of Taiwan
High Speed Rail is 315km/h
• the actual scenario: in some day, I take the train which runs in
a speed, say only 200km/h
• logical possibility: although the speed is now lower than
250km/h, it is logically possible that it is higher than 400km/h
• logical impossibility: it is logically impossible that the speed is
lower than 250km/h and higher than 400km/h simultaneously
• physical possibility: although the speed is now lower than
250km/h, it is logically possible that it is higher than 250km/h
• physical impossibility: it is physically impossible that the speed
is higher than 400km/h
FLOLAC2009@NTU 52
 Example II: Logical Possibility and Impossibility

w0
S<250
S<400

S≥250 S<250 S≥250

S<400 S≥ 400 S≥ 400
w1 w2 w3

Actually: S<250 ∧ S< 400

Logical possibility: S ≥ 250 ∧ S≥ 400
Logical impossibility: S<250 ∧ S≥ 400

FLOLAC2009@NTU 53
 Example II: Physical Possibility and Impossibility

w0
S<250
S<400

S≥250 S<250 S≥250

S<400 S≥ 400 S≥ 400
w1 w2 w3

Actually: S<250 ∧ S< 400

Physical possibility: S ≥ 250 ∧ S< 400
Physical impossibility: S ≥ 250 ∧ S≥ 400

FLOLAC2009@NTU 54
 Possibility: Logical and Physical
• actuality vs possibility:
– actuality: something actually happens, e.g., p: the speed
of the train is under 250km/h, q: the speed of the train is
under 400km/h
– possibility: although something has happened, it might have
not happened if there are other alternatives, e.g., the speed
of the train might have been higher than 250km/h if the
driver have chosen to do so, p ∧ ♦¬p
• physical possibility: under the physical constraint, the driver
does not have the alternative to make the speed of the train
higher than 315km/h
• logical possibility: however, logically, we can imagine such a
situation
FLOLAC2009@NTU 55
 Syntax and Semantics
• Basic Modal Logic—Syntax
• Basic Modal Logic—Semantics
• Basic Model Theory

FLOLAC2009@NTU 56
 Disjoint Union: Definitions and Invariance
• two models M1 = (W1, R1, π1) and M2 = (W2, R2, π2) are
disjoint if W1 ∩ W2 = ∅
• for disjoint models Mi = (Wi, Ri, πi)(i ∈ I), their disjoint
union is ]iMi = (W, R, π)
S
– W = i Wi
S
– R = i Ri
S
– π(p) = i πi(p) for each proposition letter
• for each wff ϕ, for each i ∈ I and w ∈ Wi, Mi, w ϕ iff
]iMi, w ϕ

FLOLAC2009@NTU 57
 Disjoint Union: Example

FLOLAC2009@NTU 58
 Disjoint Union: Application
• universal modality A: M, w Aϕ ⇔ for all u ∈ W , M, u ϕ
(i.e. M ϕ)
• universal modality is not definable in basic modal logic
• suppose we could, that is, there exists wff α(p) such that
M, w α(p) iff M p for any model M
• let M1 and M2 be models such that M1 p and M2 ¬p.
• for any w of M1, M1, w α(p), so M1 ] M2, w α(p).
• this implies M1 ] M2, u p for every world u of M2, thus
M2 p, a contradiction.

FLOLAC2009@NTU 59
 Generated Submodels: Definitions
• a model M0 = (W 0, R0, π 0) is a submodel of M = (W, R, π) if
W 0 ⊆ W , R0 = R ∩ (W 0 × W 0) and π 0(p) = π(p) ∩ W 0 for
every p
• M0  M: M0 = (W 0, R0, π 0) is a generated submodel of
M = (W, R, π) if M0 = (W 0, R0, π 0) is a submodel of M =
(W, R, π) and for all worlds w, if w ∈ W 0 and (w, u) ∈ R,
then u ∈ W 0

FLOLAC2009@NTU 60
 Generated Submodels: Definitions and Invariance
• for X ⊆ W , the submodel generated by X is the smallest
generated submodel whose domain contains X
• X = {w} is a singleton: rooted or point generated model, w
is the root
• if M0  M, then for each wff ϕ and w ∈ W 0, M, w ϕ iff
M0 , w ϕ
• M = ]iMi implies Mi  M for every i ∈ I

FLOLAC2009@NTU 61
 Generated Submodels: Example

FLOLAC2009@NTU 62
 Morphisms: Homomorphisms
• f : M → M0: a homomorphisms from a model M =
(W, R, π) to a model M0 = (W 0, R0, π 0) is a function f :
W → W 0 such that for w, u ∈ W and atomic proposition p
– w ∈ π(p) implies f (w) ∈ π 0(p) and
– (w, u) ∈ R implies (f (w), f (u)) ∈ R0
• M: the source and M0: the target

FLOLAC2009@NTU 63
 Homomorphisms: Example

modal formulas are not invariant under homomorphisms:

M, w ¬♦p but M0, f (w) ♦p

FLOLAC2009@NTU 64
 Morphisms: Strong Homomorphisms
• a homomorphism f : M → M0: is a strong homomorphisms if
it satisfies
– w ∈ π(p) iff f (w) ∈ π 0(p) and
– (w, u) ∈ R iff (f (w), f (u)) ∈ R0
• embedding: an injective (1-1) strong homomorphism
• isomorphism: a bijective (1-1 and onto) strong homomorphism
• if f : M → M0 is a surjective (onto) strong homomorphism,
then for each wff ϕ and w ∈ W , M, w ϕ iff M0, f (w) ϕ

FLOLAC2009@NTU 65
 Morphisms: Bounded Morphisms
• a bounded morphisms from a model M = (W, R, π) to a model
M0 = (W 0, R0, π 0) is a function f : W → W 0 such that for
w, u ∈ W , v 0 ∈ W 0, and proposition letter p
– w ∈ π(p) iff f (w) ∈ π 0(p);
– (w, u) ∈ R implies (f (w), f (u)) ∈ R0; and
– if (f (w), u0) ∈ R0 then there exists u ∈ W such that
(w, u) ∈ R and f (u) = u0 (back condition)
• f : M  M0 if f is a surjective (onto) bounded morphism
• if f : M → M0 is a bounded morphism, then for each wff ϕ
and w ∈ W , M, w ϕ iff M0, f (w) ϕ

FLOLAC2009@NTU 66
 Strong Homomorphism and Bounded Morphisms: Difference

FLOLAC2009@NTU 67
 Bisimulations: Definitions and Invariance
• Z : M↔M0: a bisimulation between M = (W, R, π) and
M0 = (W 0, R0, π 0) is a relation Z ⊆ W × W 0 such that
– if wZw0 then w ∈ π(p) iff w0 ∈ π 0(p) for any p;
– if wZw0 and (w, u) ∈ R then there exists u0 ∈ W 0 such
that uZu0 and (w0, u0) ∈ R0(forth condition); and
– if wZw0 and (w0, u0) ∈ R0 then there exists u ∈ W such
that uZu0 and (w, u) ∈ R (back condition)
• if Z : M↔M0 and wZw0 then w and w0 are bisimular, denoted
by Z : M, w↔M0, w0

FLOLAC2009@NTU 68
 Bisimulations: Definitions and Invariance
• if for some Z : M, w↔M0, w0, we write M, w↔M0, w0 or
w↔w0
• if for some Z : M↔M0, we write M↔M0
• if M, w↔M0, w0 then M, w ϕ iff M0, w0 ϕ for each wff ϕ
• the invariance can be proved by induction on the structure of
ϕ

FLOLAC2009@NTU 69
 Bisimulations and Related Notions
• if M and M0 are isomorphic, then M↔M0
• disjoint union: for every i ∈ I and w ∈ Mi, Mi, w↔ ]i Mi, w
• generated submodel: if M0  M, then M0, w↔M, w for all
w in M0
• bounded morphism: if f : M  M0, then M, w↔M0, f (w)
for all w in M

FLOLAC2009@NTU 70
 Bisimulations and Computation
• computational interpretation of a model: a process (a finite
state automaton if the model is finite)
• a possible world is a state
• the accessibility relation is simply the state transition relation
• the set of wff true in a state is the language accepted by the
automaton with the state as the initial state
• bisimulation: two model are bisimilar if they are observationally
equivalent black boxes

FLOLAC2009@NTU 71
 Bisimulations: Back and Forth Conditions

FLOLAC2009@NTU 72
 Bisimulations: Example I

Z = {(1, a), (2, b), (2, c), (3, d), (4, e), (5, e)}

FLOLAC2009@NTU 73
 Bisimulations: Example II

• bisimulation: M, s↔N, u
• modal equivalence: let Φ0 = ∅, then for each wff ϕ, M, s ϕ
iff N, u ϕ
• bisimulation implies modal equivalence
FLOLAC2009@NTU 74
 Bisimulations: Example III

• M, s and N, t are not bisimilar

• M, s (⊥ ∨ ♦⊥) and N, t 6 (⊥ ∨ ♦⊥)
• M, w ⊥ means that w is an end point
• not bisimular and not modally equivalent

FLOLAC2009@NTU 75
 Bisimulations: Example IV

• each of M and N has a finite branch of length n for each n > 0

• N has an additional infinite branch
• let Φ0 = ∅, then for each wff ϕ, M, w ϕ iff N, w0 ϕ
• however, w and w0 are not bisimular
• not bisimular but modally equivalent

FLOLAC2009@NTU 76
 Bisimulations: Hennessy-Milner Theorem
• a model M = (W, R, π) is image-finite if for all w ∈ W
R(w) = {u | (w, u) ∈ R} is finite
• w ! w0: let M = (W, R, π) and M0 = (W 0, R0, π 0) be two
models, then w ∈ W and w0 ∈ W 0 are modally equivalent if
for each wff ϕ, M, w ϕ iff M0, w0 ϕ
• Hennessy-Milner theorem: let M = (W, R, π) and M0 =
(W 0, R0, π 0) be two image-finite models, then for every w ∈ W
and w0 ∈ W 0, w↔w0 iff w ! w0

FLOLAC2009@NTU 77
 Proof of Hennessy-Milner Theorem
• w↔w0 implies w ! w0: the invariance theorem
• w ! w0 implies w↔w0: ! is itself a bisimulation
1. the first condition is immediate, so proof of forth condition
2. assume that w ! w0 and (w, u) ∈ R
3. if there is no u0 ∈ W 0 such that (w0, u0) ∈ R and u ! u0
4. let S 0 = R0(w0), then S 0 is finite and nonempty
5. S 0 = {w10 , · · · , wn0 }
6. for every wi0 ∈ S 0, there exists ψi such that M, u ψi but
M0, wi0 6 ψi
7. M, w ♦(∧iψi) but M, w0 6 ♦(∧iψi)
8. contradiction with w ! w0

FLOLAC2009@NTU 78
 Normal Systems of Modal Logic
Let us make systems out of the diamonds

FLOLAC2009@NTU 79
 Normal Systems of Modal Logic
• Hilbert-Style Axiomatization of Normal Systems
• Basic Notions of Proof Theory
• Properties of Proof-Theoretic Notions

FLOLAC2009@NTU 80
 Normal Systems of Modal Logic—Motivation

Q: given a class of models C, are there syntactic mechanisms

capable of generating the formulas valid on C?

A: the normal systems

FLOLAC2009@NTU 81
 Hilbert-Style Axiomatization—Axiom Schemata
• PL: all instances of propositional tautologies
• Dual: ♦ϕ ≡ ¬¬ϕ
• K: (ϕ ⊃ ψ) ⊃ (ϕ ⊃ ψ)
• D: ϕ ⊃ ♦ϕ
• T: ϕ ⊃ ϕ
• B: ϕ ⊃ ♦ϕ
• 4: ϕ ⊃ ϕ
• 5: ♦ϕ ⊃ ♦ϕ

FLOLAC2009@NTU 82
 Hilbert-Style Axiomatization—Rules of Inference
• R1 (Modus ponens, MP):
ϕ ϕ⊃ψ
ψ

• R2 (Generalization, Gen):
ϕ
ϕ

FLOLAC2009@NTU 83
 Remarks on Axiomatic Schemata
• PL: the starting point for modal reasoning
• K:
– distribution axiom: the distribution of  operator over ⊃
operator
– transform (ϕ ⊃ ψ) into (ϕ ⊃ ψ)
– valid in all Kripke models (prove it!)
– alethic reading: if ϕ necessarily implies ψ and ϕ is necessary,
then ψ is necessary
– epistemic reading: if an (ideal) agent knows that ϕ implies
ψ and knows ϕ, then he also knows ψ

FLOLAC2009@NTU 84
 Remarks on Axiom Schemata
• Dual: why do we need it?
– expansion of shorthand: ♦ϕ ≡ ¬¬♦¬¬ϕ
– by PL: ♦ϕ ≡ ♦¬¬ϕ
– we need it because K+PL only give us ϕ ≡ ¬¬ϕ, which
in turn give us ♦¬ϕ ≡ ♦¬¬¬ϕ (with expansion of  and
PL)

FLOLAC2009@NTU 85
 Remarks on Axiom Schemata
• D:
– alethic reading: if something is necessary, then it is possible
– deontic reading: if something is obligatory, then it is per-
mitted
• T:
– alethic reading: if something is necessary, then it is actually
true
– epistemic reading: what is known is true (verity of knowl-
edge)
– knowledge axiom or truth axiom
– distinguished feature of knowledge from belief

FLOLAC2009@NTU 86
 Remarks on Axiom Schemata
• B: what is actually true is necessarily possible
• 4:
– positive introspection axiom
– epistemic reading: if you know something, then you know
that you know it
• 5:
– it is equivalent to ¬ϕ ⊃ ¬ϕ
– negative introspection axiom
– epistemic reading: if you don’t know something, then you
know that you don’t know it

FLOLAC2009@NTU 87
 Normal Systems of Modal Logic
• the minimal normal system K: PL+Dual+K+MP+Gen
• Lemmon code for normal systems: KX0 . . . Xm−1 denote the
system K plus axiomatic schemata X0, . . ., and Xm−1
• some well-known systems
1. KT=T= the Gödel/Feys/Von Wright system
2. KT4=S4
3. KT4B=KT45=S5=epistemic system
4. KD= deontic T
5. KD4=deontic S4
6. KD45=deontic S5=doxastic system
7. KTB=the Brouwer (Brouwersche) system

FLOLAC2009@NTU 88
 Normal Systems of Modal Logic
• Hilbert-Style Axiomatization of Normal Systems
• Basic Notions of Proof Theory
• Properties of Proof-Theoretic Notions

FLOLAC2009@NTU 89
 Proof Theory: Basic Notions
• S: a normal system, Σ: a set of wffs, ϕ: a wff
• S-proof: a finite sequence of wffs, each of which is an instance
of an axiom schema in S, or follows from one or more earlier
items in the sequence by applying a rule of inference
• `S ϕ:
– ϕ is a theorem of S
– there is an S-proof ϕ0, · · · , ϕk such that ϕ = ϕk

FLOLAC2009@NTU 90
 Proof Theory: Basic Notions
• Σ `S ϕ:
– ϕ is a local syntactic consequence of ϕ in S
– there some finite subset {σ1, · · · , σn} ⊆ Σ such that `S
σ1 ∧ · · · ∧ σn ⊃ ϕ
• Σ `gS ϕ:
– ϕ is a global syntactic consequence of ϕ in S
– there is a finite sequence of wffs ϕ0, · · · , ϕk such that ϕ =
ϕk and each ϕi is
∗ an instance of an axiom schema in S
∗ an element of Σ, or
∗ follows from one or more earlier items in the sequence by
applying a rule of inference
FLOLAC2009@NTU 91
 Proof Theory: Basic Notions
• Σ is S-inconsistent: Σ `S ⊥; otherwise, Σ is S-consistent
• Σ is S-maximal: Σ is S-consistent and for any ϕ 6∈ Σ, Σ ∪ {ϕ}
is S-inconsistent
• Σ is S-closed: if Σ `S ϕ, then ϕ ∈ Σ
• Σ is an S-system: if Σ is S-closed

FLOLAC2009@NTU 92
 Proof Theory: Example
• K-proof of (p ∧ q) ⊃ (p ∧ q)
1. p ⊃ (q ⊃ (p ∧ q)) PL
2. (p ⊃ (q ⊃ (p ∧ q))) Gen : 1
3. (p ⊃ (q ⊃ (p ∧ q))) ⊃ (p ⊃ (q ⊃ (p ∧ q))) Inst. of K
4. p ⊃ (q ⊃ (p ∧ q)) MP : 2, 3
5. (q ⊃ (p ∧ q)) ⊃ (q ⊃ (p ∧ q)) Inst. of K
6. p ⊃ (q ⊃ (p ∧ q)) PL : 4, 5
7. (p ∧ q) ⊃ (p ∧ q) PL : 6

FLOLAC2009@NTU 93
 Normal Systems of Modal Logic
• Hilbert-Style Axiomatization of Normal Systems
• Basic Notions of Proof Theory
• Properties of Proof-Theoretic Notions

FLOLAC2009@NTU 94
 Local Syntactic Consequence: Properties
1. `S ϕ iff ∅ `S ϕ iff for every Σ, Σ `S ϕ
2. If Σ `PL ϕ then Σ `S ϕ
3. If ϕ ∈ Σ, then Σ `S ϕ
4. Cut: if Σ `S ψ and {ψ} `S ϕ, then Σ `S ϕ
5. If Σ `S ϕ and Σ ⊆ Γ then Γ `S ϕ
6. Compactness theorem: Σ `S ϕ iff there is a finite subset Γ of
Σ such that Γ `S ϕ
7. Deduction theorem: Σ `S ϕ ⊃ ψ iff Σ ∪ {ϕ} `S ψ

FLOLAC2009@NTU 95
 Consistency: Properties
1. Σ is S-consistent iff there is a ϕ such that Σ 6`S ϕ
2. Σ is S-consistent iff there is no ϕ such that both Σ `S ϕ and
Σ `S ¬ϕ
3. If Σ is S-consistent and Γ ⊆ Σ, then Γ is S-consistent
4. Compactness theorem: Σ is S-consistent iff every finite subset
of Σ is
5. Σ `S ϕ iff Σ ∪ {¬ϕ} is S-inconsistent
6. Σ 6`S ¬ϕ iff Σ ∪ {ϕ} is S-consistent

FLOLAC2009@NTU 96
 Maximality: Properties I

Let Σ be an S-maximal set of wffs. Then:

1. ϕ ∈ Σ iff Σ `S ϕ
2. Σ is S-closed and an S-system
3. ⊥ 6∈ Σ
4. > ∈ Σ
5. ¬ϕ ∈ Σ iff ϕ 6∈ Σ
6. ϕ ∧ ψ ∈ Σ iff both ϕ ∈ Σ and ψ ∈ Σ
7. ϕ ∨ ψ ∈ Σ iff either ϕ ∈ Σ or ψ ∈ Σ
8. ϕ ⊃ ψ ∈ Σ iff if ϕ ∈ Σ then ψ ∈ Σ
9. ϕ ≡ ψ ∈ Σ iff ϕ ∈ Σ if and only if ψ ∈ Σ
FLOLAC2009@NTU 97
 Lindenbaum Lemma
• if Σ is S-consistent, then there exists an S-maximal set of wffs
Γ such that Σ ⊆ Γ
• Γ is called an S-maximal extension of Σ

FLOLAC2009@NTU 98
 Lindenbaum Construction
1. Assume a fixed enumeration of all wffs: ϕ1, ϕ2, · · ·
2. Define Γ0 = Σ
3. For n > 0,

Γn−1 ∪ {ϕn}, if it is S-consistent
Γn =
Γn−1, otherwise
S
4. Γ = n≥0 Γn

FLOLAC2009@NTU 99
 Proof of Lindenbaum Lemma
1. for n ≥ 0, Γn is S-consistent
2. for n ≥ 0, Γn ⊆ Γ, so Σ = Γ0 ⊆ Γ
3. for n ≥ k ≥ 0, Γk ⊆ Γn
4. for n > 0, if ϕn ∈ Γ, then ϕn ∈ Γn
5. for any finite subset Γ0 of Γ, Γ0 ⊆ Γn for some n ≥ 0
6. Γ is S-consistent (by 1 and 5)
7. for every ϕ, if Γ ∪ {ϕ} is S-consistent, then ϕ ∈ Γ
8. Γ is S-maximal

FLOLAC2009@NTU 100
 Corollaries of Lindenbaum Lemma
1. Σ `S ϕ iff ϕ is in every S-maximal extension of Σ
2. `S ϕ iff ϕ is in every S-maximal set
3. | ϕ |S:
• the proof set of ϕ in S
• the set of all S-maximal sets containing ϕ

FLOLAC2009@NTU 101
 Proof Set: Properties
• Let Σ be S-maximal, then Σ ∈| ϕ |S iff ϕ ∈ Σ iff Σ `S ϕ
• | ϕ |S⊆| ψ |S iff `S ϕ ⊃ ψ
• | ϕ |S=| ψ |S iff `S ϕ ≡ ψ
• | ⊥ |S= ∅
• | > |S= the set of all S-maximal sets
• | ¬ϕ |S=| > |S − | ϕ |S
• | ϕ ∧ ψ |S=| ϕ |S ∩ | ψ |S
• | ϕ ∨ ψ |S=| ϕ |S ∪ | ψ |S

FLOLAC2009@NTU 102
 Maximality: Properties II

Let Σ and Γ be S-maximal sets of wffs. Then:

1. {ϕ | ϕ ∈ Σ} ⊆ Γ iff {♦ϕ | ϕ ∈ Γ} ⊆ Σ
2. ψ ∈ Σ iff for every S-maximal set of wffs Γ such that
{ϕ | ϕ ∈ Σ} ⊆ Γ, ψ ∈ Γ
3. ♦ψ ∈ Σ iff for some S-maximal set of wffs Γ such that
{♦ϕ | ϕ ∈ Γ} ⊆ Σ, ψ ∈ Γ

FLOLAC2009@NTU 103
 Maximality: Proof of Property II.1
• left to right
{ϕ | ϕ ∈ Σ} ⊆ Γ and ϕ ∈ Γ
⇒ ¬ϕ 6∈ Γ
⇒ ¬ϕ 6∈ Σ
⇒ ¬¬ϕ ∈ Σ
⇒ ♦ϕ ∈ Σ

• right to left
{♦ϕ | ϕ ∈ Γ} ⊆ Σ and ϕ ∈ Σ
⇒ ¬♦¬ϕ ∈ Σ
⇒ ♦¬ϕ 6∈ Σ
⇒ ¬ϕ 6∈ Γ
⇒ ϕ∈Γ
FLOLAC2009@NTU 104
 Maximality: Proof of Property II.2
• left to right: trivial
• right to left
1. Suppose that ψ is in every S-maximal extension of
{ϕ | ϕ ∈ Σ}
2. {ϕ | ϕ ∈ Σ} `S ψ (by corollary of Lindenbaum lemma)
3. there are wffs ϕ1 · · · ϕn ∈ {ϕ | ϕ ∈ Σ} such that
`S (ϕ1 ∧ · · · ∧ ϕn) ⊃ ψ (by def)
4. `S (ϕ1 ∧ · · · ∧ ϕn) ⊃ ψ (by modal reasoning)
5. Σ `S ϕi for 1 ≤ i ≤ n (by 3 and maximality of Σ)
6. Σ `S ψ (by 4,5, and modal reasoning)
7. ψ ∈ Σ (by 6 and maximality of Σ)

FLOLAC2009@NTU 105
 Maximality: Proof of Property II.3
♦ψ ∈ Σ
iff ¬¬ψ ∈ Σ (by Dual and maximality of Σ)
iff ¬ψ 6∈ Σ (by maximality of Σ)
iff for some S-maximal set of wffs Γ such that {ϕ | ϕ ∈ Σ} ⊆ Γ,
¬ψ 6∈ Γ (by Property II.2)
iff for some S-maximal set of wffs Γ such that {♦ϕ | ϕ ∈ Γ} ⊆ Σ,
¬ψ 6∈ Γ (by Property II.1)
iff for some S-maximal set of wffs Γ such that {♦ϕ | ϕ ∈ Γ} ⊆ Σ,
ψ ∈ Γ (by maximality of Γ)

FLOLAC2009@NTU 106
 Meta-Theorems of Normal Systems
Let us take a closer look at the systems

FLOLAC2009@NTU 107
 Meta-Theorems of Normal Systems
• Soundness
• Completeness
• Finite Model Property and Decidability

FLOLAC2009@NTU 108
 Classes of Models
• D: the class of all serial models
• T: the class of all reflexive models
• B: the class of all symmetric models
• 4: the class of all transitive models
• 5: the class of all Euclidean models

FLOLAC2009@NTU 109
 Correspondence between Axioms and Classes of Models

every instance of the axiom schema X is valid in the corresponding

class of models X as shown in the following table:
Schema D T B 4 5
Model Classes D T B 4 5
let us prove the correspondence between B and symmetric models
as an example ↓

FLOLAC2009@NTU 110
 Correspondence between Axioms and Classes of Models
1. let M = (W, R, π) be any symmetric model, ϕ be any wff, and
w ∈ W such that M, w ϕ
2. for any u such that (w, u) ∈ R, we have (u, w) ∈ R, so
M, u ♦ϕ
3. therefore, M, w ♦ϕ
4. M ϕ ⊃ ♦ϕ

FLOLAC2009@NTU 111
 Fifteen Distinct Normal Systems
K KD KT KB K4 K5 KDB KD4 KD5 K45
serial • • • •
reflexive •
symmetry • •
transitive • • •
Euclidean • • •

FLOLAC2009@NTU 112
 Fifteen Distinct Normal Systems
KD45 KB4 KTB KT4 KT5
serial • • •
reflexive • • • •
symmetry • • • • • •
transitive • • • • •
Euclidean • • • •

FLOLAC2009@NTU 113
 Soundness

Let S be the normal system KX1 . . . Xn (n ≥ 0) and C denote

C1 ∩ . . . ∩ Cn where each Ci is the corresponding class of models
for axiom schema Xi. Then for any subset of wffs Σ and wff ϕ,
1. if `S ϕ then C ϕ
2. if Σ `S ϕ then Σ C ϕ

FLOLAC2009@NTU 114
 Proof of Soundness
• 2 follows from 1 by definition
• proof of 1: by induction on the length of an S-proof
– every instance of axiom schemata is valid in the correspond-
ing class of models
– every application of inference rules preserves the validity

FLOLAC2009@NTU 115
 Meta-Theorems of Normal Systems
• Soundness
• Completeness
• Finite Model Property and Decidability

FLOLAC2009@NTU 116
 Completeness

Let S be the normal system KX1 . . . Xn (n ≥ 0) and C denote

C1 ∩ . . . ∩ Cn where each Ci is the corresponding class of models
for axiom schema Xi. Then for any subset of wffs Σ and wff ϕ,
1. if Σ C ϕ then Σ `S ϕ
2. if C ϕ then `S ϕ

FLOLAC2009@NTU 117
 Proof of Completeness
• suppose that Σ 6`S ϕ
• Σ ∪ {¬ϕ} is S-consistent (by property 6 of consistency)
• existence of S-maximal extension of Σ∪{¬ϕ} (by Lindenbaum
lemma)
• canonical model construction
• verify that the canonical model is in C
• Σ ∪ {¬ϕ} is S-satisfiable in the canonical model (truth lemma)
• Σ 6 C ϕ (by the definition of validity)

FLOLAC2009@NTU 118
 Canonical Model Construction

M = (W, R, π)
• W : the set of all S-maximal sets of wffs
– each world is a S-maximal set
– each world w is identified with an S-maximal set Σw
– there is a world containing Σ ∪ {¬ϕ}
• (w, u) ∈ R iff {ϕ | ϕ ∈ Σw } ⊆ Σu
• π(p) =| p |S (the proof set of p) for every atomic proposition p

FLOLAC2009@NTU 119
 Properties of the Canonical Model
• if S contains D, then M is serial
• if S contains T, then M is reflexive
• if S contains B, then M is symmetric
• if S contains 4, then M is transitive
• if S contains 5, then M is Euclidean
let us prove item 4 as an example ↓

FLOLAC2009@NTU 120
 Properties of the Canonical Model
1. suppose S contains 4, (w, u) and (u, v) ∈ R
2. ϕ ⊃ ϕ ∈ Σw (1 and property I.1 of maximality)
3. {ϕ | ϕ ∈ Σw } ⊆ Σu (1 and def. of canonical model)
4. {ϕ | ϕ ∈ Σu} ⊆ Σv (1 and def. of canonical model)
5. let us consider any ϕ ∈ Σw , then ϕ ∈ Σw
(2 and property I.8 of maximality)
6. ϕ ∈ Σu (3 and 5)
7. ϕ ∈ Σv (4 and 6)
8. (w, v) ∈ R (5, 7, and def. of canonical model)

FLOLAC2009@NTU 121
 Truth Lemma
M, w ϕ iff ϕ ∈ Σw
• by induction on the structure of ϕ
• induction base: by definition of π and properties of a proof set
• induction step for ¬ψ, ϕ1 ∧ ϕ2: exercise
• induction step for ♦ψ:
M, w ♦ψ
iff there exists u such that (w, u) ∈ R and M, u ψ (by def.
of satisfaction)
iff there exists u such that {ϕ | ϕ ∈ Σw } ⊆ Σu and ψ ∈ Σu
(by def. of R and induction hyp.)
iff for some Σu such that {♦ϕ | ϕ ∈ Σu} ⊆ Σw , ψ ∈ Σu (by
property II.1 of maximality)
iff ♦ψ ∈ Σw (by property II.3 of maximality)
FLOLAC2009@NTU 122
 Proof of Completeness
• there is a w ∈ W such that Σ ∪ {¬ϕ} ⊆ Σw
• M, w ψ for every wff ψ in Σ ∪ {¬ϕ}
• Σ ∪ {¬ϕ} is S-satisfiable in the canonical model
• Σ 6 C ϕ (by the definition of validity)

FLOLAC2009@NTU 123
 Meta-Theorems of Normal Systems
• Soundness
• Completeness
• Finite Model Property and Decidability

FLOLAC2009@NTU 124
 Finite Model Property via Filtrations
• for a wff ϕ, find the set of all its subformulas Sub(ϕ) (it’s a
finite set)
• filtrations of a model: for a (possibly infinite) model M =
(W, R, π), find a model that identifies as many worlds as pos-
sible according to Sub(ϕ) (such a filtration of M is finite)
• if ϕ is satisfiable in M, then it is also satisfiable in a filtration
of M
• ϕ is satisfiable in a model iff it is satisfiable in a finite model
• the size of such a finite model is bounded by a function of the
size of ϕ
• the satisfiability checking of ϕ is decidable by enumeration of
all models within this bounded size
FLOLAC2009@NTU 125
 Subformula Closed Sets
• a set of wffs Σ is closed under subformulas (subformula closed)
if for all wffs ϕ and ψ
– if ¬ϕ ∈ Σ then so is ϕ;
– if ♦ϕ ∈ Σ then so is ϕ; and
– if ϕ ∧ ψ ∈ Σ then so are ϕ and ψ
• Note: for a wff ϕ, Sub(ϕ) is closed under subformulas
• for a model M = (W, R, π) and a subformula closed set Σ,
define an equivalence relation Σ⊆ W × W by
w Σ u iff for all ϕ ∈ Σ: (M, w ϕ iff M, u ϕ)
• [w]Σ (or simply [w]): the equivalence class of a world with
respect to Σ
FLOLAC2009@NTU 126
 Filtrations: Definition

Let M = (W, R, π) be a model and Σ be a subformula closed set.

Suppose MfΣ is any model (W f , Rf , π f ) such that
1. W f = WΣ = {[w]Σ | w ∈ W }
2. if (w, u) ∈ R, then ([w], [u]) ∈ Rf
3. if ([w], [u]) ∈ Rf , then for all ♦ϕ ∈ Σ, if M, u ϕ then
M, w ♦ϕ
4. π f (p) = {[w] | w ∈ π(p)} for all atomic propositions p in Σ
Then MfΣ is a filtration of M through Σ.

FLOLAC2009@NTU 127
 Filtrations: Example
• M = (N, R, π)
– R = {(0, 1), (0, 2), (1, 3)} ∪ {(n, n + 1) | n ≥ 2}
– π(p) = N − {0}; π(q) = {1}
• Σ = {♦p, p}
• Σ= {(0, 0)} ∪ {(i, j) | i, j > 0}
• Mf = ({[0], [1]}, Rf , π f )
– Rf = {([0], [1]), ([1], [1])}
– π f (p) = {[1]}

FLOLAC2009@NTU 128
 Filtrations: Example

FLOLAC2009@NTU 129
 Filtration Theorem

Let MfΣ = (WΣ, Rf , π f ) be a filtration of M = (W, R, π) through

a subformula closed set Σ. Then for all wff ϕ ∈ Σ and w ∈ W ,
M, w ϕ ⇔ MfΣ, [w] ϕ
1. by induction on the structure of ϕ since Σ is subformula closed
2. induction base: by definition of π f
3. Boolean cases: straightforward
4. M, w ♦ϕ ⇒ MfΣ, [w] ♦ϕ: clause 2 of the definition of
filtration
5. MfΣ, [w] ♦ϕ ⇒ M, w ♦ϕ: clause 3 of the definition of
filtration

FLOLAC2009@NTU 130
 Existence of Filtrations

let M = (W, R, π) be a model and Σ be a subformula closed set.

1. define Rs ⊆ WΣ × WΣ:
• ([w], [u]) ∈ Rs iff there exist w0 ∈ [w] and u0 ∈ [u] such
that (w, u) ∈ R
2. define Rl ⊆ WΣ × WΣ:
• ([w], [u]) ∈ Rl iff for all ♦ϕ ∈ Σ: M, u ϕ implies M, w
♦ϕ
3. both (WΣ, Rs, π f ) and (WΣ, Rl , π f ) are filtrations of M
through Σ
4. if (WΣ, Rf , π f ) is any filtration of M through Σ then Rs ⊆
Rf ⊆ Rl
FLOLAC2009@NTU 131
 Filtrations: Remarks
• all filtrations of M preserve the reflexivity and seriality
• not all filtrations of M preserve the transitivity, symmetry, and
Euclideaness ⇒ we need some techniques to select among the
filtrations (Cf. Chellas, 1980)
• in some cases, we have to consider not only subformulas of a
wff but also its modal closure

FLOLAC2009@NTU 132
 Finite Model Property and Decidability
• a filtration of M through Σ has at most 2n possible worlds if
the cardinality of Σ is n
• if ϕ is satisfiable then it is satisfiable on a finite model contain-
ing at most 2|Sub(ϕ)| possible worlds
• consistency checking (theoremhood, validity checking, and sat-
isfiability test) in the fifteen normal systems mentioned above
is decidable

FLOLAC2009@NTU 133
 Variants of Modal Logics
Let us meet the challenge of diversity

FLOLAC2009@NTU 134
 Variants of Modal Logic
• Generalization of Basic Modal Logic
• First-order Correspondence of Modal Logic
• Multi-agent Epistemic Logic
• Dynamic Logic

FLOLAC2009@NTU 135
 Modal Logic in a More General Setting
• we need more than one relations in the relational structure
• the arity of each relation may be greater than 2
• each k-ary relation corresponds to a (k − 1)-ary modal operator
• example: in basic modal logic, the binary relation R corresponds
to the unary modal operator ♦

FLOLAC2009@NTU 136
 General Modal Logic—Syntax
• modal similarity type: τ = (∆, ρ)
– ∆: a set of modal operators {M1, M2, · · ·}
– ρ : ∆ → N the arity function
• well-formed formulas: M L(τ, Φ0)
– any atomic propositional variable is a wff
– ⊥ is a wff
– if ϕ and ψ are wffs, so are ¬ϕ and ϕ ∧ ψ
– if M∈ ∆, ρ(M) = k, and ϕ1, · · · , ϕk are wffs, so is
M (ϕ1, · · · , ϕk )

FLOLAC2009@NTU 137
 General Modal Logic—Syntax
• O(ϕ1, · · · , ϕk ): shorthand of ¬ M (¬ϕ1, · · · , ¬ϕk )
• nullary modal operators: modal constants (also propositional
constants)
• unary modal operators: Mi is usually written as ♦i, and Oi as
i
• binary modal operators: M (ϕ, ψ) is usually written as ϕ M ψ

FLOLAC2009@NTU 138
 General Modal Logic—Semantics
• τ frame: F = (W, (RM)M∈τ )
– W : a set of possible worlds (points, states, etc.)
– RM ⊆ W k+1 if ρ(M) = k
• τ model: M = (F, π)
– F: a τ frame
– π : Φ0 → 2W : a truth assignment
• when ρ(M) > 0, M, w M (ϕ1, · · · , ϕk ) ⇔
there exist u1, · · · , uk ∈ W such that (w, u1, · · · , uk ) ∈ RM
and M, ui ϕi for 1 ≤ i ≤ k
• when ρ(M) = 0, M, w M⇔ w ∈ RM

FLOLAC2009@NTU 139
 General Modal Logic—Semantics
• the difference between modal operators and Boolean connec-
tives
– Boolean connectives are truth-functional: the truth value of
a wff in a world depends on the truth values of its compo-
nents in the same world
– modal operators are non-truth-functional: the truth value of
a wff in a world depends on the truth values of its compo-
nents in the accessible worlds
• example:
– M, w ϕ ∧ ψ ⇔ M, w ϕ and M, w ψ
– M, w ϕ M ψ ⇔ there exist u, v ∈ W such that
(w, u, v) ∈ RM and M, u ϕ and M, v ψ

FLOLAC2009@NTU 140
 Example of General Modal Logic—Basic Temporal Logic
• modal similarity type: τ = (∆, ρ)
– ∆ = {F, P, X}
– ρ(F ) = ρ(P ) = ρ(X) = 1
• temporal model: M = (W, RF , RP , RX , π)
– W = Z: the set of integers (time points)
– (i, j) ∈ RF if i < j
– (i, j) ∈ RP if i > j
– (i, j) ∈ RX if j = i + 1
• M, i F ϕ ⇔ there exist j > i such that M, j ϕ
• M, i P ϕ ⇔ there exist j < i such that M, j ϕ
• M, i Xϕ ⇔ M, i + 1 ϕ
FLOLAC2009@NTU 141
 Example of General Modal Logic—Arrow Logic
• modal similarity type: τ = (∆, ρ)
– ∆ = {ι, ⊗, ◦}
– ρ(ι) = 0, ρ(⊗) = 1, ρ(◦) = 2
• arrow model: M = (W, I, R, C, π)
– W =: a set of arrows (directed arcs)
–I⊆W
– R⊆W ×W
– C ⊆W ×W ×W

FLOLAC2009@NTU 142
 Example of General Modal Logic—Arrow Logic

• M, w ι ⇔ ι ∈ I
• M, w ⊗ϕ ⇔ there exist u ∈ W such that (w, u) ∈ R and
M, u ϕ
• M, w ϕ◦ψ ⇔ there exist u, v ∈ W such that (w, u, v) ∈ C,
M, u ϕ, and M, v ψ

FLOLAC2009@NTU 143
 Example of General Modal Logic—Finite State Automata Logic

finite state automation for 0n 1m (n, m > 0)

• modal similarity type: τ = ({0, 1}, ρ), ρ(0) = ρ(1) = 1

• ♦i is written as hii
• M, s0 h0inh1imp for all n, m > 0

FLOLAC2009@NTU 144
 Example of General Modal Logic—Finite State Automata Logic

• a finite state automation: A = (Q, I, δ, s0, F )

– Q: a finite set of states
– I: a finite set of input symbols
– δ ⊆ Q × I × Q: (labeled) state transition relation
– s0 ∈ Q: initial state
– F ⊆ Q: the set of final states
• L(A) ⊆ I ∗: the language (set of strings) accepted by A

FLOLAC2009@NTU 145
 Example of General Modal Logic—Finite State Automata Logic
• modal similarity type: τ = (I, ρ) and ρ(i) = 1 for i ∈ I
• Φ0 = {p}
• MA = (Q, (Ri)i∈I , π)
– (s, t) ∈ Ri iff (s, i, t) ∈ δ
– π(p) = F
• if α = i1i2 · · · ik ∈ I ∗ then hαi = hi1ihi2i · · · hik i
• M, s0 hαip iff α ∈ L(A)

FLOLAC2009@NTU 146
 Variants of Modal Logic
• Generalization of Basic Modal Logic
• First-order Correspondence of Modal Logic
• Multi-agent Epistemic Logic
• Dynamic Logic

FLOLAC2009@NTU 147
 First Order Correspondence Language
• M L(τ, Φ0): modal language defined by the modal similarity
type τ = (∆, ρ) and the set of atomic propositions Φ0
• L1τ (Φ0): first order language with equality which has:
– unary predicates Pi for each pi ∈ Φ0
– (n + 1)-ary predicate symbol RM for M∈ ∆ if ρ(M) = n
• α(x) denote a FOL formula with one free variable x

FLOLAC2009@NTU 148
 Standard Translation from ML to FOL
• let x be a first order variable, STx : M L(τ, Φ0) → L1τ (Φ0):
1. STx(p) = P (x) if p ∈ Φ0
2. STx(⊥) = x 6= x
3. STx(¬ϕ) = ¬STx(ϕ)
4. STx(ϕ ∧ ψ) = STx(ϕ) ∧ STx(ψ)
5. STx(M (ϕ1, · · · , ϕn)) = ∃y1 . . . ∃yn(RM(x, y1, · · · , yn) ∧
STy1 (ϕ1) ∧ · · · ∧ STy1 (ϕn))

FLOLAC2009@NTU 149
 Standard Translation from ML to FOL
• Kripke model M = (W, (RM)M∈τ , π) as an FOL interpretation
– W : the domain of interpretation
– RM: the interpretation of the predicate symbol RM)
– π(p): the interpretation of the predicate symbol P
• for all M = (W, (RM)M∈τ , π) and w ∈ W
– M, w ϕ iff M STx(ϕ)[x ← w], where x ← w means
that we use a variable assignment such that w is assigned
to x
– M ϕ iff M ∀xSTx(ϕ)

FLOLAC2009@NTU 150
 Standard Translation from ML to FOL: Example

STx(♦(p ⊃ q)) = ∃y(Rxy ∧ STy (p ⊃ q))

= ∃y(Rxy ∧ (STy (p) ⊃ STy (q)))
= ∃y(Rxy ∧ (∀z(Ryz ⊃ STz (p)) ⊃ Qy))
= ∃y(Rxy ∧ (∀z(Ryz ⊃ P z) ⊃ Qy))

FLOLAC2009@NTU 151
 Importing Properties of FOL to ML
• compactness property: if Θ is a set of FOL formulas, and every
finite subset of Θ is satisfiable, then so is Θ itself.
• Löwenheim-Skolem property: if a set of FOL formulas has an
infinite model, then it has a countably infinite model
• modal logic has both these properties

FLOLAC2009@NTU 152
 Bisimulations: Example II Revisited

• bisimulation: M, s↔N, u
• modal equivalence: let Φ0 = ∅, then for each modal wff ϕ,
M, s ϕ iff N, u ϕ
• not FOL equivalent: R(x, x) is satisfiable in N (with x assigned
u) but not in M
• R(x, x) 6= STx(ϕ) for any modal wff ϕ

FLOLAC2009@NTU 153
 van Benthem Characterization Theorem
• FOL formula α(x) is invariant for bisimulation if for all
M, w↔M0, w0, M α[x ← w] iff M0 α[x ← w0]
• characterization theorem: α(x) is invariant for bisimulation iff
it is equivalent to the standard translation of a modal wff
• proof of “if” part: follows from invariance of modal wff under
bisimulation
• proof of “only if” part: beyond the scope of the course

FLOLAC2009@NTU 154
 Variants of Modal Logic
• Generalization of Basic Modal Logic
• First-order Correspondence of Modal Logic
• Multi-agent Epistemic Logic
• Dynamic Logic

FLOLAC2009@NTU 155
 Epistemic Logic—Historical Notes
• knowledge in philosophy: epistemology (starting with the
Greeks)
• formal logical analysis: [von Wright, 1951]
• Hintikka[1962]: knowledge and belief
• Fagin, Halpern, Vardi, Moses[1995]: Reasoning about knowl-
edge

FLOLAC2009@NTU 156
 Epistemic Logic—Some Applications
• computer science: AI, distributed systems, multi-agent systems,
security protocols, etc.
• linguistics: discourse reasoning
• economics: game theory

FLOLAC2009@NTU 157
 Common Knowledge
• the facts that everyone knows, everyone knows that every-
one knows, everyone knows that everyone knows that everyone
knows, and so on.
• examples:
– three wise men
– coordinated attack
– email game [Rubinstein,1989]
– mediated email game [Dimitri, 2003]

FLOLAC2009@NTU 158
 Three Wise Men

Once upon a time, a king

wished to test his three wise men.
He arranged them in a circle so
that they can see and hear each other.

FLOLAC2009@NTU 159
 Three Wise Men

The king announced that he will put a

white or black hat on each of their
heads but that at least one hat will be
white.

In fact all three hats are white.

FLOLAC2009@NTU 160
 Three Wise Men

No He repeatedly asked them, ``Do you

know the color of your hat?''

No
No

The first time, the response is ``no”

FLOLAC2009@NTU 161
 Three Wise Men

No He repeatedly asked them, ``Do you

know the color of your hat?''

No
No

The second time, the response is still

``no”

FLOLAC2009@NTU 162
 Three Wise Men

white! He repeatedly asked them, ``Do you

know the color of your hat?''

white! white!

The third time, the response is finally

``white!”

FLOLAC2009@NTU 163
 Three Wise Men
• how can it happen that the king helps the wise men along by
telling them something they already know?
• how can the wise men learn definite positive facts from hearing
statements of ignorance?
• common knowledge

FLOLAC2009@NTU 164
 Coordinated Attack

ok

Attack at dawn

each time the messenger makes it, the level of knowledge rises.
first KB m, then KAKB m, KB KAKB m, . . .

FLOLAC2009@NTU 165
 Coordinated Attack
• m will never become common knowledge using a k-round hand-
shake protocol.
• m will never become common knowledge in any run of any
protocol. In fact, common knowledge is not attainable in any
system where communication is not guaranteed.

FLOLAC2009@NTU 166
 Distributed Knowledge
• distributed knowledge is that can be deduced by pooling to-
gether the knowledge of everyone
• application: belief fusion of multiple agents, collective intelli-
gence

FLOLAC2009@NTU 167
 Logical Omniscience Problem
• modeling of ideal agents with unbounded reasoning capability
• if Kiϕ then Kiψ for any logical consequence ψ of ϕ
• example: if an agent knows the basic axioms of probability
theory, then he knows all of its theorems
• in practice, agents do not have such a magic power

FLOLAC2009@NTU 168
 Epistemic Logic—Syntax
• alphabet
– a set of atomic propositional variables Φ0 = {p1, p2, · · ·}
– a set of agents Nn = {1, 2, · · · , n}
– primitive logical symbols: ⊥, ¬, ∧, Ki (i ∈ Nn), CG, DG,
EG (G ⊆ Nn)
– defined logical symbols: >, ∨, ⊃, ≡
– auxiliary symbols: (, )
• formation rules of wffs
ϕ ::= p | ¬ϕ | ϕ ∧ ψ | Kiϕ | EGϕ | CGϕ | DGϕ
where i ∈ Nn and G ⊆ Nn

FLOLAC2009@NTU 169
 Epistemic Logic—Intuition
• Kiϕ: agent i knows (or believes) ϕ
• EGϕ: every agent in G knows ϕ
• DGϕ: ϕ is distributed knowledge for the agent group G
• CGϕ: ϕ is common knowledge for the agent group G

FLOLAC2009@NTU 170
 Epistemic Logic—Semantics

M = (W, (Ri)1≤i≤n, π)
• W : a set of possible worlds (points)
• Ri: a binary relation over W for each i (the epistemic alterna-
tive relation)
• π : Φ0 → 2W
• DG = ∩i∈GRi
• EG = ∪i∈GRi
• CG = ∪k≥0EGk = (∪i∈GRi)∗

FLOLAC2009@NTU 171
 Epistemic Logic—Semantics
• M, w |= p iff w ∈ π(p)
• M, w |= ¬ϕ iff M, w 6|= ϕ
• M, w |= ϕ ∧ ψ iff M, w |= ϕ and (M, w |= ψ
• M, w |= Kiϕ iff M, u |= ϕ for all (w, u) ∈ Ri
• M, w |= EGϕ iff M, u |= ϕ for all (w, u) ∈ EG
• M, w |= CGϕ iff M, u |= ϕ for all (w, u) ∈ CG
• M, w |= DGϕ iff M, u |= ϕ for all (w, u) ∈ DG

FLOLAC2009@NTU 172
 Epistemic Logic—Semantics
• agent i in the world w may know something about the world,
but does not know what the world is exactly
• (w, u) ∈ Ri means that the agent considers that the actual
world may be u while he is actually in the world w
• Ri(w) = {u | (w, u) ∈ Ri} is the set of all worlds that agent
i consider possible while he is actually in w
• i’s knowledge about w constrains the worlds he considers pos-
sible
• if he is totally ignorant of the world w, Ri(w) = W
• if he knows ϕ is true, those worlds satisfying ¬ϕ will be excluded
from Ri(w)

FLOLAC2009@NTU 173
 Epistemic Logic—Semantics
• distributed knowledge: when a group of agents can pool their
knowledge together, their constraints on Ri(w) are also pooled
together
• a world is considered possible by the group, only when all agents
in the group consider it possible, so DG = ∩i∈GRi

FLOLAC2009@NTU 174
 Epistemic Logic—Semantics
• everybody knows: the knowledge that everybody knows will
constrain Ri(w) of each i
• if everybody knows ϕ, only worlds satisfying ϕ will remain in
Ri(w) for each agent i
• if everybody knows ϕ, only worlds satisfying ϕ will remain in
∪i∈GRi(w), so EG = ∪i∈GRi
• common knowledge: “everybody knows” depends on EG, “ev-
erybody knows everybody knows” depends on EG2 , and so no,
so CG = ∪k≥0EGk

FLOLAC2009@NTU 175
 Distributed Knowledge: Example

1 5 15 7 11 13

student 1 2 4 0 6 12 8 10 14

student 2

The teacher announce to two students that he writes down a

natural number 0≤ n ≤ 15 in a paper.
He told student 1 that the number is even privately.
He told student 2 that the number is divisible by 3 privately.
The distributed knowledge of both students: the number is 0 or 6 or 12

FLOLAC2009@NTU 176
 The Kripke Model of Three Wise Men

(1,1,0) (1,1,1)

(1,0,0) (1,0,1)
•Initial condition
•actual situation: (0,0,0)
(0,1,0) (0,1,1)
•all possible situations
•Agent 1:
•Agent 2:
•Agent 3:
•0: white, 1: black

(0,0,0) (0,0,1)

FLOLAC2009@NTU 177
 The Kripke Model of Three Wise Men

K3white3
(1,1,0)
after the king’s announcement ,
the situation that
all hats are black is deleted

K2white2

(1,0,0) (1,0,1)
K1white1
(0,1,0) (0,1,1)

(0,0,0) (0,0,1)

FLOLAC2009@NTU 178
 The Kripke Model of Three Wise Men

since no response to the first call:

delete all possible worlds where
some agent knows the color of his hat
K2white2 ∧ K3white3

(1,0,0)
K1white1 ∧ K3white3
(0,1,0)

K1white1 ∧ K2white2

(0,0,0) (0,0,1)

FLOLAC2009@NTU 179
 The Kripke Model of Three Wise Men

since no response to the second call:

delete all possible worlds where
some agent knows the color of his hat,
in the only remaining world, every agent knows
that the color of his hat is white

K1white1 ∧ K2white2 ∧ K3white3

(0,0,0)

FLOLAC2009@NTU 180
 Knowledge in Multi-agent System
• global state: (se, s1, · · · , sn)
• local state: si is the local state of agent i and se is the state
of the environment
• run (execution): a function from time (natural numbers) to
global states
• a system R: a set of runs
• point: a pair (r, m) consisting of a run r and a time m (r(m)
is a global state)
• if r(m) = (se, s1, · · · , sn), then take ri(m) to be si

FLOLAC2009@NTU 181
 A System

FLOLAC2009@NTU 182
 A Instance of Kripke Model—Interpreted System Semantics
• interpreted system: I = (R, π)
– R: a system (a set of runs)
– π: π(p) is a set of global states
• for each 1 ≤ i ≤ n, Ki is an equivalence relation on points:
(r, m)Ki(r0, m0) iff ri(m) = ri0 (m0)

• DG = ∩i∈GKi
• EG = ∪i∈GKi
• CG = (∪i∈GKi)∗

FLOLAC2009@NTU 183
 An Instance of Kripke Models—Interpreted System Semantics
• (I, r, m) |= p iff r(m) ∈ π(p)
• (I, r, m) |= ¬ϕ iff (I, r, m) 6|= ϕ
• (I, r, m) |= ϕ ∧ ψ iff (I, r, m) |= ϕ and (I, r, m) |= ψ
• (I, r, m) |= Kiϕ iff (I, r0, m0) |= ϕ for all (r0, m0) ∈ Ki(r, m)
• (I, r, m) |= EGϕ iff (I, r0, m0) |= ϕ for all (r0, m0) ∈ EG(r, m)
• (I, r, m) |= CGϕ iff (I, r0, m0) |= ϕ for all (r0, m0) ∈ CG(r, m)
• (I, r, m) |= DGϕ iff (I, r0, m0) |= ϕ for all (r0, m0) ∈ DG(r, m)

FLOLAC2009@NTU 184
 An Instance of Kripke Models—Interpreted System Semantics
• a point (r, m) corresponds to a possible world in a Kripke model
• Ki corresponds to Ri
• each Ki is determined by the local state of agent i
• agent i can know only his local state, so he considers (r, m)
possible while he is in (r0, m0) (i.e. (r, m) ∈ Ki(r0, m0)), if his
local state is the same in both (r, m) and (r0, m0)
• each Ki is an equivalence relation

FLOLAC2009@NTU 185
 Model Classes in Epistemic Logic
• Cn: all models
• Crn all reflexive models
• Crt
n all reflexive and transitive models

• Crst
n all equivalence models

• Crlt
n all Euclidean, serial, and transitive models

FLOLAC2009@NTU 186
 Axiom Schemata
• PL: all tautologies of the propositional calculus
• K: (Kiϕ ∧ Ki(ϕ ⊃ ψ)) ⊃ Kiψ
• T: Kiϕ ⊃ ϕ
• 4: Kiϕ ⊃ KiKiϕ
• 5: ¬Kiϕ ⊃ Ki¬Kiϕ
• D: ¬Ki⊥
V
• C1: EGϕ ≡ i∈G Ki ϕ
• C2: CGϕ ⊃ EG(ϕ ∧ CGϕ)
• D1: D{i}ϕ ≡ Kiϕ
• D2: DGϕ ⊃ DG0 ϕ if G ⊆ G0

FLOLAC2009@NTU 187
 Rules of Inference
• R1 (Modus ponens, MP):
ϕ ϕ⊃ψ
ψ

• R2 (Knowledge Generalization, Gen):

ϕ
Ki ϕ

• RC1 (Induction):
ϕ ⊃ EG(ψ ∧ ϕ)
ϕ ⊃ CG ψ

FLOLAC2009@NTU 188
 Hilbert Style Axiomatic Systems
• Kn: PL, K, MP, Gen
• Tn: Kn+T
• S4n: Kn+T+4
• S5n: Kn+T+4+5
• KD45n: Kn+4+5+D
• SC : S+C1+C2+RC1
• SD : S+D1+D2
• SCD : SC +SD

FLOLAC2009@NTU 189
 Completeness
• the correspondence:
axiom constraint
PL, K none
T reflexive
4 transitive
5 Euclidean
D serial
rst
• example: `S5CD
n
ϕ iff ϕ is valid in C n .

FLOLAC2009@NTU 190
 Complexity
• knowledge and common knowledge:
Logic Completeness
S51, KD451 NP
Kn, Tn, S4n, n ≥ 1; P SP ACE
S5n, KD45n, n ≥ 2
KCn , TCn , n ≥ 1; EXP T IM E
S4Cn , S5Cn , KD45Cn , n ≥ 2
• adding distributed knowledge to the language does not affect
complexity

FLOLAC2009@NTU 191
 Variants of Modal Logic
• Generalization of Basic Modal Logic
• First-order Correspondence of Modal Logic
• Multi-agent Epistemic Logic
• Dynamic Logic

FLOLAC2009@NTU 192
 Dynamic Logic: Reasoning about Programs
• program: a recipe written in a formal language for computing
desired output data from given input data
• programs typically use variables to hold input and output values
and intermediate results
• state: a function that assigns a value to each program variable
• a program can be viewed as a transformation on states
• proposition: a description of a state

FLOLAC2009@NTU 193
 Dynamic Logic: Programming Constructs
• programs are built inductively from atomic programs and tests
using various program operators
• regular program operators:
– sequential composition: α; β
– non-deterministic choice: α ∪ β
– iteration: α∗

FLOLAC2009@NTU 194
 Dynamic Logic: Program Verification
• I/O specification: ϕ, ψ
• partial correctness: whenever a program started in a state sat-
isfying the input condition ϕ, then if it halts, it does so in a
state satisfying the output condition ψ
• total correctness: partially correct and halts whenever it started
in a state satisfying ϕ
• dynamic logic vs temporal logic:
– dynamic logic is exogenous: programs are explicit in the
logical language
– temporal logic is endogenous: program is fixed and is consid-
ered part of the structure over which the logic is interpreted

FLOLAC2009@NTU 195
 Precursor of Dynamic Logic: Hoare Logic
• {ϕ}α{ψ}: α is partially correct with respect to the I/O speci-
fication ϕ, ψ (FOL formulas)
• rules
– assignment rule: if e is free for x in ϕ
{ϕ[x/e]}x := e{ϕ}

– composition rule::
{ϕ}α{ψ}, {ψ}β{χ}
{ϕ}α; β{χ}

FLOLAC2009@NTU 196
 Precursor of Dynamic Logic: Hoare Logic
• rules
– conditional rule:
{ϕ ∧ ψ}α{χ}, {ϕ ∧ ¬ψ}β{χ}
{ϕ}if ψ then α else β{χ}

– while rule:
{ϕ ∧ ψ}α{ψ}
{ψ}while ϕ do α{¬ϕ ∧ ψ}

– weakening rule:
ϕ0 ⊃ ϕ {ϕ}α{ψ} ψ ⊃ ψ 0
{ϕ0}α{ψ 0}

FLOLAC2009@NTU 197
 Propositional Dynamic Logic: Syntax
• alphabet
– a set of atomic propositional variables Φ0 = {p1, p2, · · ·}
– a set of atomic programs Π0 = {a1, a2, · · ·}
– propositional operators: ⊥, ¬, ∧
– program operators: ; (sequential composition), ∪ (non-
deterministic choice), ∗ (iteration)
– mixed operators: [ ] (necessity modality formation operator),
?(test)
– defined symbols: >, ∨, ⊃, ≡, hi (possibility modality for-
mation operator)
– auxiliary symbols: (, )

FLOLAC2009@NTU 198
 Propositional Dynamic Logic: Syntax
• wffs Φ and programs Π: the smallest sets such that
– Φ0 ⊆ Φ and Π0 ⊆ Π
– if ϕ, ψ ∈ Φ, then ⊥, ¬ϕ, ϕ ∧ ψ ∈ Φ
– if α, β ∈ Π, then α; β, α ∪ β, α∗ ∈ Π
– if α ∈ Π and ϕ ∈ Φ, then [α]ϕ ∈ Φ
– if ϕ ∈ Φ, then ϕ? ∈ Π
• abbreviation: hαiϕ = ¬[α]¬ϕ
• precedence of operators: unary operators bind tighter than bi-
nary ones, and ; binds tighter that ∪

FLOLAC2009@NTU 199
 Propositional Dynamic Logic: Syntax
• [α]ϕ: it is necessary that after executing α, ϕ is true
• α; β: execute α, then execute β
• α ∪ β: choose either α or β non-deterministically and execute
it
• α∗: execute α non-deterministically chosen finite number of
times (zero or more)
• ϕ?: proceed if ϕ is true, fail if false

FLOLAC2009@NTU 200
 Propositional Dynamic Logic: Shorthand
• skip: >?
• fail: ⊥?
• if ϕ1 → α1 | · · · | ϕn → αn fi: ϕ1?; α1 ∪ · · · ∪ ϕn?; αn
• do ϕ1 → α1 | · · · | ϕn → αn od:
(ϕ1?; α1 ∪ · · · ∪ ϕn?; αn)∗; (¬ϕ1 ∧ · · · ∧ ¬ϕn)?
• if ϕ then α else β: if ϕ → α | ¬ϕ → β fi
• while ϕ do α: do ϕ → α od
• repeat α until ϕ: α; while ¬ϕ do α
• {ϕ}α{ψ}: ϕ ⊃ [α]ψ

FLOLAC2009@NTU 201
 Propositional Dynamic Logic—Semantics
• (Kripke) model: M = (W, (Ra)a∈Π0 , π)
– W : a set of possible worlds (points, states, etc.)
– Ra ⊆ W × W : a binary relation over W
– π : Φ0 → 2W : a truth assignment
• intuition
– a model is a state transition system
– Ra: the set of I/O pairs of states of the atomic program a

FLOLAC2009@NTU 202
 Propositional Dynamic Logic—Semantics

given a model M = (W, (Ra)a∈Π0 , π) and w ∈ W

• state transition of compound programs:
– Rα;β = Rα ◦ Rβ =
{(w, u) | ∃v ∈ W, (w, v) ∈ Rα , (v, u) ∈ Rβ }
– Rα∪β = Rα ∪ Rβ
– Rα∗ = n≥0 Rαn
S

– Rϕ? = {(w, w) | M, w ϕ}
• satisfaction:
– M, w [α]ϕ ⇔
for every u ∈ W , if (w, u) ∈ Rα , then M, u ϕ

FLOLAC2009@NTU 203
 Propositional Dynamic Logic—Semantics
• satisfiability and validity are defined as in basic modal logic
• the set of all finite computation sequences of α: CS(α)
– CS(a) = {a} if a ∈ Π0
– CS(ϕ?) = {ϕ?}
– CS(α; β) = {γδ | γ ∈ CS(α), δ ∈ CS(β)}
– CS(α ∪ β) = CS(α) ∪ CS(β)
– CS(α∗) = n≥0 CS(αn), where α0 = skip and αk+1 =
S
α; αk
S
• property: Rα = σ∈CS(α) Rσ

FLOLAC2009@NTU 204
 PDL Semantics: Example

• M p ≡ [(ab∗a)∗]p (we write α; β as αβ)

• M q ≡ [(ba∗b)∗]q
• α = (aa ∪ bb ∪ (ab ∪ ba)(aa ∪ bb)∗(ab ∪ ba))∗
• M ϕ ≡ [α]ϕ for any ϕ

FLOLAC2009@NTU 205
 PDL—Axiom Schemata
1. PL: all tautologies of the propositional calculus
2. [α](ϕ ⊃ ψ) ⊃ ([α]ϕ ⊃ [α]ψ)
3. [α ∪ β]ϕ ≡ ([α]ϕ ∧ [β]ϕ)
4. [α; β]ϕ ≡ [α][β]ϕ
5. [ϕ?]ψ ≡ (ϕ ⊃ ψ)
6. (ϕ ∧ [α][α∗]ϕ) ≡ [α∗]ϕ
7. induction axiom: ϕ ∧ [α∗](ϕ ⊃ [α]ϕ) ⊃ [α∗]ϕ

FLOLAC2009@NTU 206
 PDL—Rules of Inference
• R1 (Modus ponens, MP):
ϕ ϕ⊃ψ
ψ

• R2 (Generalization, Gen):
ϕ
[α]ϕ

FLOLAC2009@NTU 207
 PDL—Alternative Formulations of the Induction Axiom
• RTC (reflexive transitive closure rule):
ϕ ⊃ (ψ ∧ [α]ϕ)
ϕ ⊃ [α∗]ψ

• LI (loop invariance rule):

ϕ ⊃ [α]ϕ
ϕ ⊃ [α∗]ϕ

• IND (the induction axiom): ϕ ∧ [α∗](ϕ ⊃ [α]ϕ) ⊃ [α∗]ϕ

• IND♦ (in dual form): hα∗iϕ ⊃ (ϕ ∨ hα∗i(¬ϕ ∧ hαiϕ))

FLOLAC2009@NTU 208
 PDL—Meta-Theorems
• soundness and completeness: let ϕ be a PDL wff, then we have
ϕ iff ` ϕ
• the satisfiability problem for PDL is EXPTIME-complete
• RTC, LI, IND, and IND♦ are inter-derivable in PDL without
the induction axiom
• the rules of Hoare logic are derivable in PDL
• compactness fails for PDL

FLOLAC2009@NTU 209
 Inter-derivation of Induction Axioms
• (IND) ⇒(LI):
1. premise of LI: ϕ ⊃ [α]ϕ
2. [α∗](ϕ ⊃ [α]ϕ) (Gen,1)
3. ϕ ⊃ (ϕ ∧ [α∗](ϕ ⊃ [α]ϕ)) (2. PL)
4. ϕ ⊃ [α∗]ϕ (IND, PL, 3)
• (LI)⇒(RTC):
1. premise of RTC: ϕ ⊃ (ψ ∧ [α]ϕ)
2. ϕ ⊃ ψ (PL, 1)
3. ϕ ⊃ [α]ϕ (PL, 1)
4. ϕ ⊃ [α∗]ϕ (LI, 3)
5. [α∗](ϕ ⊃ ψ) (Gen, 2)
6. ϕ ⊃ [α∗]ψ (Ax2, PL, 4, 5)
FLOLAC2009@NTU 210
 Inter-derivation of Induction Axioms
• (RTC)⇒(IND):
1. ϕ ∧ [α∗](ϕ ⊃ [α]ϕ)
⊃ ϕ ∧ (ϕ ⊃ [α]ϕ) ∧ [α][α∗](ϕ ⊃ [α]ϕ) (Ax6)
2. ϕ ∧ [α∗](ϕ ⊃ [α]ϕ)
⊃ ϕ ∧ [α]ϕ ∧ [α][α∗](ϕ ⊃ [α]ϕ) (1, PL)
3. ϕ ∧ [α∗](ϕ ⊃ [α]ϕ)
⊃ ϕ ∧ [α](ϕ ∧ [α∗](ϕ ⊃ [α]ϕ)) (2, modal reasoning)
4. ϕ ∧ [α∗](ϕ ⊃ [α]ϕ) ⊃ [α∗]ϕ (3, RTC)
• (IND♦) ⇔(IND): PL and the duality of [α] and hαi

FLOLAC2009@NTU 211
 PDL Encoding of Hoare Logic
• we derive the while rule:
{ϕ ∧ ψ}α{ψ}
{ψ}while ϕ do α{¬ϕ ∧ ψ}
1. premise: (ϕ ∧ ψ) ⊃ [α]ψ
2. ψ ⊃ (ϕ ⊃ [α]ψ) (PL,1)
3. ψ ⊃ [ϕ?; α]ψ (Ax4, Ax5, 2)
4. ψ ⊃ [(ϕ?; α)∗]ψ (LI rule, 3)
5. ψ ⊃ [(ϕ?; α)∗](¬ϕ ⊃ (¬ϕ ∧ ψ)) (PL, Ax2, MP, 4)
6. ψ ⊃ [(ϕ?; α)∗; ¬ϕ?](¬ϕ ∧ ψ)) (Ax4, Ax5, 5)

FLOLAC2009@NTU 212
 Failure of Compactness
• let Σ = {p ⊃ q, p ⊃ [a]q, p ⊃ [a2]q, · · ·}
• ϕ = p ⊃ [a∗]q
• Σ ϕ in PDL: for every model M and w, M, w Σ implies
M, w ϕ
• if Σ0 ⊂ Σ and p ⊃ [ak ]q 6∈ Σ0 for some k, then M0, 0 Σ0 but
M0, 0 6 ϕ

FLOLAC2009@NTU 213
 PDL—Remarks
• our finite state automata logic is a special case of PDL without
program operators (only atomic programs are allowed)
• note the analogy between CG in multi-agent epistemic logic and
[α∗] in PDL
• translation from multi-agent epistemic logic KCn to PDL: me2d
1. me2d(p) = p
2. me2d(⊥) = ⊥
3. me2d(¬ϕ) = ¬me2d(ϕ)
4. me2d(ϕ ∧ ψ) = me2d(ϕ) ∧ me2d(ψ)
5. me2d(Kiϕ) = [i]me2d(ϕ)
6. me2d(CGϕ) = [ G][( G)∗]me2d(ϕ)
S S

FLOLAC2009@NTU 214
 Conclusion
You can now explore the modal space by yourself

FLOLAC2009@NTU 215
 Conclusion
• The entry point to the modal space
• Summary of the course
• A rough guide to the future study

FLOLAC2009@NTU 216
 The Entry Point
• you have arrived at the entry point of the modal space
• you are ready to answer the following questions posed by the
“immigration officer”
1. what is the purpose of your visit?—why would you like to
study modal logic?
2. where will you stay?—which part of modal logic you are
interested in?
3. how long will you stay?—will your interests in modal logic
persist?
• after all, you should be able to form a coherent map of the
territory
• but, before that, I will provide a rough guide to you
FLOLAC2009@NTU 217
 Conclusion
• The entry point to the modal space
• Summary of the course
• A rough guide to the future study

FLOLAC2009@NTU 218
 Summary

we have touched on the following topics

• the evolution of modal logic
• basic modal logic:
– syntax of the diamond language
– Kripke frames and models
– basic notions of model theory: validity, satisfiability, se-
mantic consequence, disjoint union, generated submodel,
bounded morphism, and bisimulation
• normal systems of basic modal logic:
– Hilbert-style axiomatization
– proof-theoretic notions: theoremhood, consistency, syntac-
tic consequence, Lindenbaum Lemma
FLOLAC2009@NTU 219
 Summary

we have touched on the following topics

• meta-theorems of normal systems:
– soundness
– strong completeness (by canonical model construction)
– decidability by the filtration method
• variants of modal logic:
– general modal logic: more modalities and higher arity
– standard translation to FOL
– reasoning about knowledge: multi-agent epistemic logic
– reasoning about regular programs: propositional dynamic
logic
FLOLAC2009@NTU 220
 Conclusion
• The entry point to the modal space
• Summary of the course
• A rough guide to the future study

FLOLAC2009@NTU 221
 Where to Stay—Interested Areas
• computer science
– software engineering
– artificial intelligence
– world wide web
• philosophy
– formal epistemology
– philosophy of information
• mathematics
• economics and social science
• linguistics
• system science
FLOLAC2009@NTU 222
 What to See—Interesting Topics: Temporal Logics
• software engineering: software/hardware specification and ver-
ification
• basic temporal logic extended with operators S (since) and U
(until)
• model checking
• linear time vs. branching time
• point-based vs interval-based
• combination with other modal logic: temporal-epistemic and
spatio-temporal logics

FLOLAC2009@NTU 223
 What to See—Interesting Topics: Spatial Logics
• mathematics: geometry and topology
• point-based logics: logic of elsewhere and everywhere, collinear-
ity and qualitative distance
• line-based logics: logics of parallelism, orthogonality, intersec-
tions of lines
• incidence logic: incidence relation between a point and a line,
projective geometry, affine geometry
• topological logic:  as interior and ♦ as closure

FLOLAC2009@NTU 224
 What to See—Interesting Topics: Deontic Logics
• AI: normative agent systems
• computer security: specification of security policies
• standard deontic logic (SDL): accessibility relation points to
“ideal” or “perfect deontic alternatives” of the world under
consideration
• w Oϕ: ϕ is true in all such ideal worlds
• SDL suffers from a number of paradoxes
• Ross’sparadox: Oϕ ⊃ O(ϕ∨ψ): if one ought to mail the letter
then one ought to mail it or burn it
• dynamic deontic logic: Oα = [α]V : an action is obligatory if
failing to do it leads to a state of violation
• defeasible deontic logic: deontic rules with exceptions
FLOLAC2009@NTU 225
 What to See—Interesting Topics: Intelligent Agents
• AI: modeling of mental attitudes of intelligent agents
• types of mental attitudes
– informational: knowledge, belief, and awareness
– motivational: commitment, choice, intentions, plans (inter-
nal commitment), goals (a degree of internal commitment),
desire, want, wish, preference
– social: obligation and permission
– emotional: joy, hope, sorrow, happiness, fear, distress, pride,
relief, love, hate, anger, shame, gratitude etc.
• BDI logic: belief, desire, intention

FLOLAC2009@NTU 226
 What to See—Interesting Topics: Description Logics
• AI: knowledge representation
• WWW: web 3.0 (semantic web), ontology representation and
resource description language
• the alphabet: concept names (atomic concepts A, B) and role
names (atomic roles R)
• concept terms:
C ::= A | ⊥ | ¬C | C u D | ∀R : C | ∃R : C

• ∃R : C and ∀R : C correspond to hRiC and [R]C in multi-

modal logic respectively
• more expressive extensions

FLOLAC2009@NTU 227
 What to See—Interesting Topics: Boolean Modal Logics
• instances: temporal logic, dynamic logic, and multi-agent epis-
temic logic
• modal similarity type (∆, ρ), where ∆ is not only a set but also
an algebraic structure
• in temporal logic: past and future operators are mutually con-
verse
• in dynamic logic: program operators form a regular algebra
• in epistemic logic CG corresponds to a transitive closure of EG
• AI: analysis of information systems, rough set theory in data
mining

FLOLAC2009@NTU 228
 What to See—Interesting Topics: Dynamic Epistemic Logic
• applications to philosophy of information
• AI and formal epistemology: belief revision
• epistemic action: to change mental states of cognitive agents
• public announce logic: [ϕ]ψ: after the announcement of ϕ, ψ
holds
• dynamic doxastic logic: modal logic of belief revision by
Segerberg

FLOLAC2009@NTU 229
 What to See—Interesting Topics: Many-Dimensional Modal Logics
• each possible world has some inner structure instead of an ab-
stract entity: a tuple or a sequence over some base set
• the accessibility relations are (partly) determined by this inner
structure of the states
• system science: complex combined systems out of relatively
simple ones
• instances: interval temporal logic (an interval as a pair of time
points), arrow logic in square frame (an arrow as a pair of
points), combination of epistemic and temporal logics
• combinations of modal logics:
– fusion (independent join or dovetailing): combined compo-
nents do not interact at all
– product: interaction is strong
FLOLAC2009@NTU 230
 What to See—Interesting Topics: Much More
• mathematics: logic of provability, logic of justification (Arte-
mov)
• economics and social science: game logic, coalition logic, inter-
active epistemology (Aumann)
• system science: regular equivalence in social network analysis,
modeling of complex networks, different graded modalities for
uncertainty reasoning
• linguistics: feature logic, Montague semantics, hybrid logic
• philosophical logic: first-order and higher-order modal logic

FLOLAC2009@NTU 231
 Getting Around—Approaches
• syntactic approach: natural deduction, Gentzen systems,
tableau methods (in particular for description logics), resolu-
tion, translation to FOL
• semantic approach: frame correspondence theory, fragments of
FOL or HOL, model theory
• computational approach: computability and complexity, modal
logic programming
• algebraic approach: algebraic semantics for modal logic, algebra
and coalgebra
• topological approach: topological interpretation of modal logic

FLOLAC2009@NTU 232
 Side Trips—Related Logics
• conditional logic:
– original motivation of modal logic by Lewis
– possible world semantics
– ϕ → ψ: if minimal change to ϕ-world, then ψ
– related to belief revision and dynamic doxastic logic
• nonmonotonic logic: Σ ` ϕ does not imply Σ ∪ {ψ} ` ϕ
• relevant logic and substructural logic
• intuitionistic logic: Kripke semantics for intuitionistic logic
• quantum logic: Kripke semantics with Hilbert space models

FLOLAC2009@NTU 233
 Ok! My guide will stop here!

FLOLAC2009@NTU 234
 Epilogue
• The set of slides was originally designed for a mini-course (9
hours) of the FLOLAC’2009 held in National Taiwan University
• I have revised it slightly based on the response (or silence) of
the students. As in the example of three wise men, a teacher
must learn from hearing statements of ignorance, as well as
from silence.

FLOLAC2009@NTU 235
 Acknowledgements
• Most materials are from standard textbooks:
– Introduction, basic modal logic: (Blackburn et al., 2001)
– Normal systems and meta-theorems: (Chellas, 1980)
– Generalization of modal logic: (Blackburn et al., 2001)
– multi-agent epistemic logic: (Fagin et.al., 1995)
– dynamic logic: (Harel et. al., 2000)
• A few slides are from Areces and Blackburn’s ESSLLI’08 course
“Logic for Computation”
• A few slides are from Pacuit’s Stanford university course “An
Invitation to Modal Logic”
• Most pictures are downloaded from the WWW.

FLOLAC2009@NTU 236