Professional Documents
Culture Documents
Varzi
1 — INTRODUCTION
◆ Basic ideas:
— Modal logic is an extension of (not an alternative to) ordinary logic.
— It is concerned with logical facts (e.g., inferences) that involve modalities, i.e., qualifications of
propositions.
— On the other hand, the following statement is not just true: it is necessarily true (could not have
been false)
Neil is Neil.
You might want to say that this is just a logical fact: self-identities are logically true (true under
every interpretation of the non-logical constants). Then consider
Neil is a person.
Hesperus is Phosphorus.
Nothing is in two places at the same time.
These are not true in every logically possible model (i.e., under every way of interpreting our
language). Yet, arguably, they are true in every possible world. (E.g., since ‘Hesperus’ and
‘Phosphorus’ pick out one and the same object, that object could not be different from itself.)
1
◆ The basic picture (mostly from ARISTOTLE, De Interpretatione, chapters 12-13)
true false
necessary
po impossible
ss
ib
e l
contingent
2. Two questions
(I would be happy to change the lable to Theory of modality, or something like that.)
◆ Couldn’t we deal with modalities without invoking anything else than standard logic?
Four possible strategies:
2
ƒ1 yields p⇔p (determinism)
ƒ2 yields p ⇔ ¬p
ƒ3 violates p⇒p and makes a trivial operator anyhow
ƒ4 validates p ⇒ ¬p " "
In general, from the T of p we don’t know anything about the T of p
In general, from the F of p we don’t know anything about the F of p
• But Montague 1963 shows this is no good—cp. Tarski’s problems with truth predicate.
• Only recently this strategy has been reconsidered: see Schwartz 1992.
3
3. Need for a logic (Theory) of modality
◆ Necessity and possibility are alethic modalities (modifications of truth = aletheia), but there are also
— Epistemic, deontic, temporal, spatial modalities.
— Indeterminacy
— Provability
• A ⇔ A is provable (relative to some one formal system, e.g., Peano Arithmetic)
• Gödel’s second incompleteness theorem would read: ¬ ⊥ → ¬ ¬ ⊥
4
Modal Logic (G4424) Achille C. Varzi
◆ Vocabulary:
— atomic formulas: P 0, P 1, P 2, . . .
— connectives: ⊥, T, ¬, ∧, ∨, →, ↔, , .
— metavariables: A, B, C, .....
◆ Grammar:
— Straightforward.
— Only be careful to distinguish necessity of the consequence vs necessity of the consequent.
1) (A→Β)
2) A→ Β
Obviously different:
2. Semantics
1
— This induces an corresponding assignment of values to all sentences—a valuation:
Vα (A) → {T,F} for all sentences A
or equivalently
=α A iff A is true (holds, etc.) relative to α.
◆ How do we specify truth conditions for modal formulas (given that , are not truth-functional)?
1) Carnap in Meaning and Necessity:
=α A iff =β Α for every model β
=α A iff =β Α for some model β
— This is too strict: it equates worlds with models, hence necessity and logical validity.
— This account still requires that we specify what possible worlds are (functions or sets of
sentences, etc.), but otherwise OK. It can be generalized as follows:
2
— Formally: = 〈W,P〉, where
• W ≠ Ø (the possible worlds)
• P a sequence P0 , P1 , P2 , . . . ⊆ W associating with each i a set of worlds (those in
which P i holds)
• Intuitively: Pi = the proposition expressed by P i
— Truth conditions:
=α P i iff α∈Pi
M M M
=α A iff =β A for every β∈W
=α A iff =β A for some β∈W
— Notes:
• Pi may be empty
• i Pi may not add up to W
◆ Second idea (Kripke’s standard models): Generalize Leibnizian models by relativizing possibilities:
— Model = 〈W, R, P〉, where
• W and P as before
• R⊆W ×W
[Intuitively: R is a relation of “accessibility”: αRβ means that β is possible relative to α]
— Truth conditions:
=α P i iff α∈Pi
M M M
=α A iff =β A for all β∈W such that αRβ
=α A iff =β A for some β∈W such that αRβ
◆ Third idea (Montague’s minimal models): Necessity and possibility should not be understood in
terms of truth in every/some world, but treated as primitive:
— Model = 〈W, N, P〉, where
• W and P as before
• N : W → ℘℘W associates each world with the propositions that are necessary at that world
℘W = sets of worlds = propositions
℘℘W= set of propositions
[Intuitively: Nα = the propositions that are necessary at α]
3
— Truth conditions:
=α P i iff α∈Pi
M M M
=α A iff {β∈W: =β A} ∈ Nα = iff A expresses a necessary proposition at α
=α A iff {β∈W: ≠β A} ∉ Nα = iff A does not express an impossible proposition at α
3. Examples
◆ Definitions
◆ Some principles that are valid in the semantics based on Leibnizian models:
D A→ A
T A→A
B A→ A
4 A→ A
5 A→ A
G A→ A
K (A → B) → ( A → B)
Df A ↔ ¬ ¬A
Df A ↔ ¬ ¬A
RN =C A
____
=C A
RE =C A ↔ B
__________
=C A ↔ B
RK =C (A1 ∧ ... ∧ An ) → A
_____________________
=C ( A1 ∧ ... ∧ An ) → A
4
◆ Examples of invalid principles:
( Α ∧ B) → (A ∧ B) same countermodel
( Α→ B) → (A → B) same countermodel
Example1 τ( P n → P n ) = τ( P n ) → τ(P n )
= ∀xτ(P n ) → τ(P n )
= ∀xP n (x) → P n (x)
Example2 τ( (P 1 → P 2 ) → ( P 1 → P 2 )) = τ( (P 1 → P 2 )) → τ( P 1 → P 2 )
= τ( (P 1 → P 2 )) → τ( P 1 ) → τ( P 2 )
= ∀xτ(P 1 → P 2 ) → ∀xτ(P 1 ) → ∀xτ(P 2 )
= ∀x(τ(P 1 ) → τ(P 2 )) → ∀xτ(P 1 ) → ∀xτ(P 2 )
= ∀x(P 1 (x) → P 2 (x)) → (∀xP 1 (x) → ∀xP 2 (x))
5
◆ Remark: τ is a specification of the truth conditions
Proof. ⇐ If not = A, then there is a model =〈W,P〉 and a world α∈W such that not ≠α A
But =〈W,P〉 is also a model for quantificational logic.
By induction, show that for any α∈W, ≠α A iff ≠ τ(A) [sα], where sα (x) = α for all α∈W
– A atomic. Then τ(A) = P n (x) for some n. Then: ≠α A ⇔ ≠α P n
⇔ α∉Pn
⇔ sα(x)∉Pn
⇔ ≠ P n (x) [sα]
⇔ ≠ τ(P n ) [sα]
– A= B. Then τ(A) = ∀xτ(B). Then: ≠α A ⇔ ≠α B
⇔ ≠β B for some β∈W
⇔ ≠ τ(B) [sβ] (by IH)
⇔ ≠ τ(B) [sα(x|β)].
⇔ ≠ ∀ατ(B) [sα]
⇔ ≠ τ(A) [sα]
⇒ Similar Proof
6
Modal Logic (G4424) Achille C. Varzi
◆ The general notion of a model (see § 2.2): = 〈W, ..., P〉, where
• W ≠ Ø (the possible worlds)
• P a sequence P0 , P1 , P2 , . . . ⊆ W associating each i with a set of worlds (those in which P i holds)
Df A ↔ ¬ ¬Α valid valid
Df A ↔ ¬ ¬Α valid valid
RE =C A ↔ B valid valid
__________
=C A ↔ B
1
schema standard minimal
α β
α β
α β γ
P
¬◆ P
P
◆P
β
¬■ ◆ P
α P
¬◆ P
G A→ A valid iff incestual: αRβ & αRγ ⇒ ∃δ(βRδ & γRδ) (g) p. 225
■P
◆ ■ P β P
¬■ ◆ P
α ¬◆ P δ
The notation is important: the relevant conditions on R can be expressed by means of first-order
formulas, so we are essentially looking for ordinary first order models satisfying these formulas.
2
◆ Example of proof for the “if” part (from Theorem 3.5):
Scheme 5 is valid in the class of all euclidean standard models:
1. Assume =α A.
2. Then =β A for some β∈W such that αRβ.
3. Suppose αRγ.
4. Then, we have αRγ & αRβ and therefore, by euclideanness, γRβ (of course, we also have αRβ
& αRγ and therefore βRγ, but this is irrelevant).
5. So, for any γ such that αRγ there exists β such that γRβ and =β A.
6. So, for any γ such that αRγ, =γ A.
7. Thus =α A.
8. By 1–7, if =α A then =α A.
9. Hence =α A→ A.
2. General comparisons
3
◆ FACT 2 (using parts of Theorem 7.9 p. 221)
Every standard model s
= 〈W s, R, Ps〉 is pointwise equivalent to a minimal model
m
= 〈W , N, P 〉, where X∈Nα iff X contains all R-accessible worlds, i.e., iff {β∈W: αRβ} ⊆ X.
m m
(Intuitively: the propositions necessary at α are those that include the set of all worlds accessible
from α)
Proof: Again, we set ƒ(α) = α and prove by induction that, for every sentence A:
=α A iff =α A
s m
for every α∈W:
Again, the only interesting case is modal sentences:
=α =β A for all β∈W s.t. αRβ
s s
A iff
{β∈W: αRβ} ⊆ {β∈W: =β A}
s
iff
{β∈W: αRβ} ⊆ {β∈W: =β A}
m
iff by I.H.
{β∈W: =β A} ∈ Nα
m
m
iff by def. of
=α A
m
iff by recursive clause for
3. Comments
◆ Morals:
— The simplified models may be viewed as a special case of standard models
— Standard models may be viewed as a special case of minimal models
◆ Note: Fact 2 can be improved by saying under what conditions a minimal model is equivalent to a
standard model (Theorem 7.9).
4
Modal Logic (G4424) Achille C. Varzi
G A→ A ■P
◆■P β
P
R must be incestual (offprints of a common parent ¬ ■◆P
if n=0 φn A = A
if n = k+1 φn A = φφk A
1
◆ DEFINITION (p. 88)
A standard model = 〈W, R, P〉, is k,l,m,n-incestual iff αRk β & αRm γ ⇒ ∃δ(βRl δ & γRn δ)
R
k β R
l
α δ
m n
R R
γ
So in particular:
(The proofs of these equivalences se are just derivations in first order logic with identity.)
2
◆ FACT (Theorem 3.8):
The schema Gk , l , m , n is valid in the class of all k,l,m,n-incestual standard models
2. Further generalizations
Gc A→ A
Gr ( A → A) → A
Sahl n
(A → B) (with resrtictions on the form of A and B)
◆ QUESTION 1:
Does every modal formula correspond to some first-order definable R?
i.e., given a formula A, is there always a first-order sentence φ so that, for every = 〈W, R, P〉
ANSWER IS NO
3
◆ QUESTION 2:
What about the other way around? Does every R correspond to a modal formula?
ANSWER IS NO
— E.g. Reflexivity ∀α(αRα) ⇒ A→A
Irreflexivity ∀α(¬αRα) ⇒ no characteristic wff
i.e., if a wff is true in every irreflexive model, then it
is true in every model
— Ditto for
Asymmetry ∀α(αRβ→ ¬βRα)
Antisymmetry ∀α(αRβ & βRα → α=β)
Intransitivity ∀α(αRβ & βRγ → ¬αRγ)
⇒ Exercises 3.56–57
4
Modal Logic (G4424) Achille C. Varzi
◆ DEFINITION 2.11
A system of modal logic is any set of sentences (=theory) closed under the rule:
A 1 , …, An
__________
RPL (n ≥ 0) where A is a tautological consequence of A1 … An
A
◆ Equivalently: A system of modal logic is any set of sentences Σ closed under the rules:
__________
PL where A is a tautology
A
A, A→ B
__________
MP
B
◆ NOTE
— The set of all tautologies (PC) is the smallest system of modal logic
• every system closed under RPL contains all tautologies (by n=0)
— The set of all sentences valid in a class of models {Α: =
| C A}
M
— The set of all sentences valid in a model {Α: = | A}
— The set of all sentences true at a world {Α: =| Mα A}
— The set of all sentences is the largest system of modal logic
◆ DEFINITION 2.12
A theorem of a system modal logic Σ is any member of Σ:
Σ A ⇔ Α∈Σ
◆ DEFINITION 2.14
A is deducible from Γ in system Σ (Γ Σ A) iff there are A1 , …, An ∈Γ so that Σ (A1 ∧ … ∧ An ) → A
1
This is equivalent to the usual definition: Γ Σ A iff there is finite sequence A1 , …, An such that
• A1 ∈ Σ∪Γ
• Ai ∈ Σ∪Γ or follows by RPC
• An = A
◆ DEFINITION 2.15
A set Γ is inconsistent in Σ (CønΣ Γ) iff Γ Σ
◆ THEOREM 2.16
These notions have all the expected properties
◆ EXERCISES
3.33
3.35 4.5–6 read only
3.37 (seriality done in class) 2.32 (b)–(e) 4.7 (some)
3.51 4.9
3.56–7 read only
◆ DEFINITION 4.1
A system of modal logic is normal iff it contains every instance of
Df A ↔ ¬ ¬A
(A1 ∧ … ∧ An ) → A
________________________
RK (n ≥ 0)
( A1 ∧ … ∧ An ) → A
2
◆ Eventually:
complete
Classical Systems ←→ Minimal Models
sound
◆ THEOREM 4.2:
Every normal system of modal logic has the following rules of inference:
A
____
RN n=0
A
A →B
________
RM n=1
A→ B
(A ∧ B) → C
___________________
RR n=2
( A ∧ B) → C
A↔ B
________
RE RM × 2
A↔ B
N T T PL (tautol)
T RN
C ( A ∧ B) → (A ∧ B) (A ∧ B) → (A ∧ B) PL
( A ∧ B) → (A ∧ B) RR
M (A ∧ B) → ( A ∧ B) (A ∧ B) → A PL
(A ∧ B) → B PL
(A ∧ B) → A RM
(A ∧ B) → B RM
(A ∧ B) → ( A ∧ B) PL
R (A ∧ B) ↔ ( A ∧ B) (by C and M)
K (A → B) → ( A → B) ((A → B) ∧ A) → B PL
( (A → B) ∧ A) → B RR
(A → B) → ( A → B) PL
3
◆ THEOREM 4.4: We also have the analogues for .
Df A ↔ ¬ ¬A ¬A ↔ ¬ ¬¬A Df
A ↔ ¬¬A PL
A ↔ ¬¬A RE
A ↔ ¬ ¬A PL (1,3)
A → (A1 ∨ … ∨ An )
________________________
RK
A → ( A1 ∨ … ∨ An ) A → (A1 ∨ … ∨ An ) hyp
(¬A1 ∧ … ∧ ¬An ) → ¬A RPL
( ¬A1 ∧ … ∧ ¬An ) → ¬A RK
¬ ¬A → (¬ ¬A1 ∨ … ∨ ¬ ¬An ) RPL
A → ( A1 ∨ … ∨ An ) RPL + Df
Then the derived rules and theorems:
¬A
______
RN n=0
¬ A
A →B
________
RM n=1
A→ B
A → (B ∨ C)
___________________
RR n=2
A → ( B ∨ C)
A ↔B
________
RE RM × 2
A↔ B
N ¬ ⊥
C (A ∨ B) → ( A ∨ B)
M ( A ∨ B) → (A ∨ B)
R (A ∨ B) ↔ ( A ∨ B)
4
◆ THEOREM 4.6
We also have the following generalizations
(A1 ∧ … ∧ An ) → A
________________________
RKk
( k
A1 ∧ … ∧ k
An ) → k
A (A1 ∧ … ∧ An ) → A
( k –1
A1 ∧ … ∧ k –1
An ) → k –1
A I.H.
( k –1
A1 ∧ … ∧ k –1
An ) → k –1
A RK
( k
A1 ∧ … ∧ k
An ) → k
A Def
Df k k
A↔ ¬ k
¬A k –1
A ↔ ¬ k –1 ¬A I.H.
k –1
A ↔ ¬ k –1 ¬A RE
k –1
¬A ↔ ¬ ¬ k –1 ¬A Df
A↔ ¬
k –1 k –1
¬A PL
k
A ↔ ¬ k ¬A Def
k k
This can be extended to every other principle as well: The result of putting and for and
in a valid principle remains valid.
3. Alternative Characterizations
Df + RK ⇔ Df + K + RN
Df + N + RR
Df + N,C + RM
Df + N,C,M + RE
n=0 RK = RN
n=1 RK = RM A→ B
(A → B) RN
A→ B K, PL
n=k+1 (A1 ∧ … ∧ An ) → A
(A1 ∧ … ∧ An–1 ) → (An → A) PL
( A1 ∧ … ∧ An–1 ) → (An → A) I. H.
( A1 ∧ … ∧ An–1 ) → ( An → A) K, PL
( A1 ∧ … ∧ An ) → A PL
5
◆ THEOREM 4.4:
Axiomatizations using
Df + RK ⇔ Df + RK
Df + N + RR
Df + N + RR
Df + N ,C + RM
Df + N ,C ,M + RE
◆ THEOREM 4.7: Every normal system of modal logic has the following Rule of Replacement:
B ↔ B'
________
REP
A ↔ A' where A' is obtained from A by replacing ≥0 occurrences of B by B'
1. (A → B) ↔ (¬A ∨ B) PL
2. (A → B) ↔ (¬A ∨ B) 1,PL, REP
3. (A → B) ↔ ( ¬A ∨ B) 2, R , REP
4. (A → B) ↔ (¬ ¬A → B) 3, PL, REP
5. (A → B) ↔ ( A → B) 4, Df , REP
◆ EXERCISE 4.19
Df + RK ⇔ Df + K,N + REP
◆ DEFINITION 2.4, P . 30
*
Ïn = ¬ Ïn
*
T =⊥
*
⊥ =T
* *
(¬Α) = ¬(A )
* * *
(Α ∧ Β) = A ∨ B
* * *
(Α ∨ Β) = A ∧ B
6
* * *
(Α → Β) = ¬(A ) ∧ B
* * *
(Α ↔ Β) = A ↔ ¬(B )
* *
( Α) = (A )
* *
( Α) = (A )
◆ EXAMPLES
* * *
N ( T) = (T ) = ⊥
* *
C (( Α ∧ Β) → (Α ∧ Β))
* *
= ¬( Α ∧ Β) ∧ ( (Α ∧ Β))
* * *
= ¬(( Α) ∨ ( Β) ) ∧ (Α ∧ Β)
* * * *
= ¬( Α ∨ Β ) ∧ (Α ∨ Β )
◆ FACT (see THEOREM 4.8): The following equivalencies hold in any normal system Σ:
*
(1) |– Σ Α ⇔ |– Σ ¬Α proof by induction on the complexity of A
**
(2) |– Σ Α ⇔ |– Σ Α from (1)
* *
(3) |– Σ (Α → Β) ⇔ |– Σ Β → Α
Proof of (3):
* *
|– Σ Α → Β ⇒ |– Σ ¬Α → ¬Β from (1) by REP
* *
⇒ |– Σ Β → Α
* * * *
|– Σ Β → Α ⇒ |– Σ ¬Α → ¬Β from (1) by REP
* * * *
⇒ |– Σ ¬(¬Α ) → ¬ (¬Β ) from (1) by REP
** **
⇒ |– Σ Α → Β def
⇒ |– Σ Α → Β from (2) by REP
◆ DEFINITION (P . 29): the dual of a modality φ is the modality φ* obtained by interchanging and
◆ THEOREM 4.10
|– Σ φΑ → ψΑ ⇔ |– Σ ψ*Α → φ*Α
7
Modal Logic (G4424) Achille C. Varzi
D A→ A
T A→A
B A→ A
4 A→ A
5 A→ A
— Naming conventions:
KS1 ... Sn
is the (smallest) extension of K obtained by taking the schemas S1 ... Sn as axioms.
(The order of the Si does not matter.)
— E.g., KT5 is the smallest system of modal logic obtained by adding T and 5
Etc.
◆ Facts:
— There are 25 =32 possible combinations
— Only 15 of these are distinct (proved later)
— General picture: Figure 4.1 on p. 132.
◆ Example:
KTD = KT
Proof: Obviously
KT ⊆ KTD.
So we only show that
KTD ⊆ KT.
1
To this end it is sufficient to show that every instance of D is a theorem of KT:
1. KT A→A T
2. KT A→ A duality principle
3. KT A→ A 1,4 PL
◆ Other examples:
KT5 = KTD5 = KTB5 = KT45 = KTDB5 = KTD45 = KTB45 = KTDB45 (This is S5)
— Every instance of D is a theorem of KT5: obvious from above
— Every instance of B is a theorem of KT5:
1. KT5 A→ A 5
2. KT5 A→ A dual of T
3. KT A→ A 1,2 PL
— Every instance of 4 is a theorem of KT5:
1. KT5 A→ A 5
2. KT5 A→ A 5 (duality principle — Theorem 4.10)
3. KT5 A→ A 2, RM
4. KT5 A→ A B (which is a theorem of KT5)
5. KT5 A→ A 3,4, PL
2. Some other important Normal Systems (and their positions in the diagram)
G A→ A
Gc A→ A
Gr ( A → A) → A
◆ Facts about G:
KG ⊆ KB (EXERCISE 4.34)
1. KB A→ A B
2. KB A→A B (Theorem 4.10)
3. KB A→ A 1,2 PL
2
KG ⊆ K5 (EXERCISE 4.38b)
1. K5 A→ A 5
2. K5 A→ A 5 (Theorem 4.10)
3. K5 ( A→ A) → ( A→ A) theorem of K (Exercise 4.7p, p. 123)
4. K5 A→ A 2,3, PL
5. K5 A→ A 1,4 PL
K4 ⊆ KGr
KT4 ⊆/ KGr
KGr ⊆/ KT5
◆ Definition: two modalities φ and ψ are equivalent (in system Σ) iff for all sentences
Σ φA ↔ ψA
a) KT5 A↔ A 1. KT5 A→ A T
2. KT5 A→ A 4
3. KT5 A↔ A 1,2, PL
b) KT5 A↔ A 1. KT5 A→ A 4
2. KT5 A→ A T
3. KT5 A↔ A 1,2, PL
3
c) KT5 A↔ A 1. KT5 A→ A T
2. KT5 A→ A 5
3. KT5 A↔ A 1,2, PL
d) KT5 A↔ A 1. KT5 A→ A T
2. KT5 A→ A 5
3. KT5 A↔ A 1,2, PL
◆ Example:
KT5 ¬ ¬ A ↔ KT5 ¬ ¬ ¬¬ A Df
↔ KT5 ¬ ¬ A PL + REP
↔ KT5 A Df
↔ KT5 A d) above
↔ KT5 A a) above
↔ KT5 A c) above
↔ KT5 A d) above
↔ KT5 A a) above
↔ KT5 A c) above
↔ KT5 A d) above
◆ In fact, you can just drop all modalities except for the last (plus negation, if necessary)
◆ Remarks:
1) These reduction laws fix an upper bound; a lower bound (to the effect that there are no further
reduction laws) follows from completeness (Chapter 5).
2) Only 7 of the 15 basic systems in the picture have finitely many distinct modalities:
KT4 K5 KD5 K45 KB4 KD45 KT5
(See THEOREMS 4.22–4.28 for proofs)
3) Two systems may have the same modalities, but differ with respect to the patterns of
implication among them
e.g. KT5 (Theorem 4.28) six modalities
KD45 (Theorem 4.27) same six modalities, but T not provable
4) Of course, two distinct systems may not have the same modalities and the same the patterns of
implication among them
4
4. Soundness (§ 5.1)
◆ THEOREM 5.1:
Let S1 ... Sn be schemas valid in classes of standard models C1 ...Cn (respectively). Then the
system KS1 ... Sn is sound with respect to the class C1 ∩ ... ∩ Cn , i.e.,
KS 1 ... S n A ⇒ C1 ∩ ... ∩ Cn A
◆ In general, systems are sound with rspect to models as per table 5.1 (p. 164)
Proof: Find counterexamples of the sort considered in soundness, i.e. show that φA ↔ ψA fails in
a class of models with respect to which the system is sound.
5
◆ Infinitely many modalities (THEOREM 5.4):
The 8 systems
K KD KT KB K4 KDB KD4 KTB
have infinitely many modalities
The following is a reflexive symmetric model, hence a model of KTB by Theorems 3.5 + 5.1.
P0 P0 P0 ... P0 ¬P 0 ... ¬P 0
α0 α1 α2 αm αm + 1 αn
m n
We have α0 P but not α0 P
Therefore not C
m
P↔ n
P (where C is the class of reflexive, symmetric models)
Therefore not KTB
m
P↔ n
P (by soundness)
Since all of K, KD, KB, KT, KDB are included in KTB, this completes case 1.
6
Modal Logic (G4424) Achille C. Varzi
◆ We have seen how to prove soundness. Now we deal with completeness. Main strategy:
1) Define canonical standard model for Σ
2) Show: if is canonical, then for all A and all α∈W:
M
α A iff A ∈α
3) Using this, show in general: if is canonical, then
M
A iff Σ A
4) Thus, to show that Σ is complete with respect to a class C of models, it is sufficient to show
that C contains a canonical model for Σ. For then we have:
M
C A ⇒ A ⇔ Σ A
◆ THEOREM 2.16
Basic properties of Σ and Σ-consistency. Notably:
(6) Γ Σ A and Γ⊆∆, then ∆ Σ A
(14) Γ Σ A iff Γ∪{¬A} is not Σ-consistent
∴ Γ / Σ A iff Γ∪{¬A} is Σ-consistent
1
◆ DEFINITION 2.17 (p. 53)
A set of sentences Γ is Σ-maximal (written MaxΣΓ) iff
1) Γ is Σ-consistent
2) Γ has no Σ-consistent extensions, i.e., for every A, if Γ∪{A} is Σ-consistent, then A∈Γ
◆ THEOREM 2.18
Basic facts about Σ-maximal sets of sentences Γ. Notably:
1) A∈Γ iff Γ Σ A
2) Σ⊆Γ
M
5) ¬A∈Γ iff A∉Γ
6) A∧B∈Γ iff both A∈Γ and B∈Γ
M
Proof:
(⇒) Let ∆⊇Γ be Σ-maximal:
Γ Σ A ⇒ ∆ ΣA by THEOREM 2.16 (6)
⇒ A∈∆ by maximality (=THEOREM 2.18 (1))
(⇐) by contraposition:
Γ /Σ A ⇒ Γ ∪ {¬A} is Σ-consistent by THEOREM 2.16 (14)
⇒ ∃ Σ-maximal ∆⊇Γ ∪ {¬A} by THEOREM 2.19 (Lindenbaum’s)
⇒ ∃ Σ-maximal ∆⊇Γ such that A∉∆ by THEOREM 2.18 (5)
2
3. Proving completeness
◆ Recall the idea of a Henkin model: put the language in the domain
Proof by induction:
M
α Ïn iff α∈Pn by def of Mα
iff Ïn ∈α by def of canonical model
M
α ¬A iff / Mα A by def
iff A∉α by I.H.
iff ¬A∈α by maximality (THEOREM 2.18 (5))
M
M
α A iff Mβ A for all β such that αRβ by def
iff A∈β for all β such that αRβ by I.H.
iff A∈α by def of canonical model
3
STEP 3: DETERMINATION WITH RESPECT TO CANONICAL MODELS
Proof:
M
A | Mα A
iff = for all α∈W by def.
iff A ∈α for all α∈W by step 2 above
iff α Σ A for all α∈W by maximality of α (THEOREM 2.18(1))
iff Σ A from COROLLARY 2.20 (2)
◆ NOTE: This means that if is a canonical standard model for Σ, then Σ is sound and complete
with respect to the class { }
STEP 4: CONCLUSION
◆ To show:
Σ is complete with respect to a class C of models,
it is sufficient to show:
C contains a canonical model for Σ.
For then we have:
M
C A ⇒ A ⇔ Σ A
◆ In general (though not always), one shows that C contains the proper canonical standard model:
4
◆ Must show: proper canonical standard models are indeed canonical standard models, i.e., 3)
implies the corresponding clause in the definition of canonical model
◆ THEOREM 5.11
If is the proper canonical standard model for Σ, then
A∈α ⇔ A∈β for every β∈W such that αRβ
◆ THEOREM 5.12
K is determined by (=sound and complete with respect to) the class C of all standard models
Soundness: trivial
1. Df is valid in C
2. validity in C is preserved by RK
Completeness: also trivial
Let = the proper canonical standard model for K
Then is a standard model
Hence ∈C
5
◆ LEMMA 5.13
If is the proper canonical standard model for Σ, then
(1) is serial if Σ contains (every instance of) D
(2) is reflexive ´´ T
(3) is symmetric ´´ B
(4) is transitive ´´ 4
(5) is euclidean ´´ 5
In general (THEOREM 5.17):
(6) is k,l,m,n-incestual ´´ Gk,l,m,n
Proof of (4)
Show: B∈γ
◆ THEOREM 5.14
KD is determined by the class C ={ : is serial}
KT C ={ : is reflexive}
KB C ={ : is symmetric}
M M
KDB C ={ : is serial & symmetric}
M M
In general:
KS1 ... Sn is determined by the class C1 ∩ ... ∩ Cn
where S1 ... Sn ∈ {G k,l,m,n : k,l,m,n ≥ 0} and C1 ...Cn are the corresponding classes of standard
models (as per LEMMA 5.13).
6
5. Remarks
◆ In general, determination for the basic systems holds also if we take only finite models.
(This is good for decidability)
◆ Not every determination result can be established using proper canonical models.
There exist systems that are determined by a class C even if the canonical model ∉C.
E.g., the canonical model for KGr is not irreflexive, but Gr fails for any model in which R is not
irreflexive.
7
Modal Logic (G4424) Achille C. Varzi
1. Introduction
— Deontic
ÉA = It ought to be the case that A (often written °A)
— Epistemic
ÉA = The agent, x, believes that A (often written BA)
ÉA = The agent, x, knows that A (often written KA)
— Temporal
ÉA = It will always be the case that A (often written GA)
ÉA = It has always been the case that A (often written HA)
◆ É is often written ° (for Obligatory) and ® (i.e., ¬É¬) is written P (for Permissible). So:
D °A → PA (‘Ought’ implies ‘can’)
— Intuitively, these principles express different thoughts, so their equivalence is a defect of any K-
system, hence of any modal logic which admits of a Kripke-style semantics.
1
— In other words, to avoid this result we must go “below K”, hence work with a weaker
Montague-style semantics.
◆ Equivalently:
{β: αRβ} = the proposition that represents the standards of obligation for the world α.
Thus:
M M
6=α °A ⇔ 6 =β A for every β such that αRβ
M
⇔ {β: αRβ} ⊆ {β: 6=β A}
⇔ {β: αRβ} ⊆ || A ||M
⇔ the proposition expressed by A is entailed by the standard of obligation for α.
β1
γ1
α β2 γ2
γ3
β3
2
4. Looking for extensions (KD systems)
P
0
¬OP
0
(D*1) (D*2)
KD4 KD44 c KD4U
(D*) (D*3)
KD KD45
3
5. Problems with these theories (all KD systems)
◆ Adequacy:
— Cannot express conditional obligations
4
(i)–(iii) seem a reasonable and consistent set of requirements. Yet the fact that John does not go
to help his neighbors, i.e., (iv), is enough to yield a contradiction. Formally:
(1) °H given
(2) °(H → T) given
(3) ¬Η → ° ¬Τ given
(4) ¬Η given
(5) °(H → T) → (°H → °T) K
(6) °T (1), (2), (5), RPL
(7) ° ¬Τ (3), (4), RPL
(8) ° T ∧ ° ¬Τ (6), (7), RPL
(9) ¬(°T ∧ °¬T) Equivalent to D
(10) ⊥ (8), (9), RPL
6. A weaker system
A→B
________
RM
°A → °B
OD ¬ °(A ∧ ¬A)
N °(A ∨ ¬A)
C (°A ∧ °B) → °(A ∧ B)
5
◆ By correctness problem 1) (“obligations always exist”), we want to get rid of N
— But this is a K- theorem.
— This means we need a system weaker than K, hence not complete with respect to Kripke models.
— We need minimal models
7. Even weaker?
— RM implies that
¢D PA → P(A∨B).
1. ¬(A∨B) → ¬A PL
2. °¬(A∨B) → °¬A 1, RM
3. ¬°¬A → ¬°¬(A∨B) 2, PL
4. PA → P(A∨B). DfP
But this is counterintuitive:
6
— In fact, it seems natural to suppose that
Peter may drink either water or whiskey → Peter may drink water and he may drink whiskey
◆ Åkvist puzzle.
— Consider the epistemic operator Peter knows that, written K. Since knowledge implies truth,
¢ KA → Α
RM implies that
¢D °KA → °Α
— But this is counterintuitive:
Peter ought to know that there is a fire →
/ There ought to be a fire
◆ Conclusion: D is also too strong...
7
Modal Logic (G4424) Achille C. Varzi
◆ Starting point:
D BA → ¬B¬A (coherence)
4 BA → BBA (positive introspection)
5 ¬BA → B¬BA (negative introspection)
◆ Semantics
— possible worlds = possible representations (consistent and complete) of reality
— αRβ iff β is epistemically possible (= conceivable) for the agent in α
M
— 6 =α BA ⇔ x thinks that Α is ungiveupable (=a constant element of all of representations)
◆ Determination
— R is serial, transitive, euclidean. So, standard situation looks like this:
α δ
1
— Note:
M
not 6=α BA → A so T fails: beliefs need not be true
M
6=α B(BA → A) so U holds: beliefs are believed to be true
◆ Problems
— RK implies closure of beliefs under logical implication ⇒ full (implicit?) belief
To avoid this, one must go for minimal models (non-normal systems)
2. Adding Knowledge
◆ Notation:
KA =df the agent, x, knows that A
◆ This can be defined in terms of B if we accept the principle that knowledge is true belief:
DfK KA ↔ BA ∧ A.
D BA → ¬B¬A (coherence)
TK KA → A
?1 KA → BA
4K KA → KKA (introspection)
?2 BA → KBA (introspection)
?3 ¬BA → K¬BA (introspection)
?4 (BA ∧ A) → KA
RN 6 ____
=Kmix A
6=Kmix KA
2
◆ Theorems:
¢ Kmix KA ↔ BA ∧ A (=DfK)
¢ Kmix BA ↔ BKA
¢ Kmix BA ↔ ¬K¬KA
T KA → A
4 KA → KKA
5– (BA ∧ A) → KA
A → (BA → KA)
A → (¬K¬KA → KA)
A → (®ÉA → ÉA)
BA ↔ ¬K¬KA or KA ↔ BA ∧ A
◆ Other theories
BA ↔ BKA
3
2. KT5 is not good if BA ↔ ¬K¬KA
For otherwise
4
Modal Logic (G4424) Achille C. Varzi
1. Temporal logic
◆ Modalities:
FA A
FA it will sometime be the case that A
GA it will always be the case that A
= ¬F¬A
A PA
PA it has sometime be the case that A
HA it has always be the case that A
= ¬P¬A
— Axioms:
System K for G
+ System K for H
+ A → GPA
+ A → HFA
— Theorems:
¢ PGA → A
¢ FHA → A 1. ¬A → GP¬A ax
2. ¬GP¬A → A PL
3. F¬P¬A → A dfF
4. FHA → A dfH
— More generally:
¢kt A ⇔ ¢kt A*, where A* is the mirror image of A
(replace G / H and F / P)
— This means symmetry past/future
◆ Semantics:
— Note: a multimodal system
— in general: one R for each modality
1
◆ Determination: all standard models
— provided αRGβ ⇔ βRHα
— alternatively: same R in two directions (the direction of time)
◆ Natural extension:
A A
FA ∧ FB PA ∧ PB
¬F(A ∧ B) ¬P(A ∧ B)
branching: ¬F(A ∧ FB) ¬P(A ∧ PB)
¬F(FA ∧ Β) ¬P(PA ∧ Β)
B B
2. Linear extensions
CL + D GA → FA seriality
HA → PA "
2
◆ System PL: dense time (Prior)
Kt + 4 ® + GA → A
GA → HA
3. Branching extensions
◆ System CR (Cocchiarella)
Kt + 4® (= CL minus linearity)
3
Modal Logic (G4424) Achille C. Varzi
1. Preliminaries
◆ Vocabulary:
— variables x0 , x1 , x2 , ...
— names a0 , a1 , a2 , ...
— predicates œ0 , œ1 , œ2 , ...
— connectives and quantifiers ∀, ∃,
◆ Grammar:
— Straightforward
— But notice that the interplay between quantifiers and modalities yields de dicto/de re readings:
de dicto: É ∀xA e.g., Necessarily, everything is spatio-temporally located.
de re: ∀x ÉA e.g., Everything is necessarily spatio-temporally located.
de dicto: ® ∀xA e.g., It is possible that everybody votes for Berlusconi.
de re: ∀x ®A e.g., Anybody could vote for Berlusconi.
de dicto: É ∃xA e.g., There must necessarily be right-winged politicians.
de re: ∃x ÉA e.g., Somebody is necessarily a right-winged politician.
de dicto: ® ∃xA e.g., Bush might have had a sister.
de re: ∃x ®A e.g., Somebody might have been Bush’s sister.
◆ Semantics:
— Two main approaches.
1. The possibilist (or fixed-domain) approach: each model comes with a single domain of
quantification containing all possible objects (i.e., all objects that are possible according to
the model).
2. The actualist (or world-relative) approach: each model comes with a domain of quantifica-
tion for each world in the model—a domain containing only those objects that actually exist
in the given world.
— We shall look at these options only from the perspective of Kripke-style semantics, but
Montague-style semantics admit of a similar distinction.
1
2. Semantics # 1: The fixed-domain approach
◆ MODELS
A model is a four-tuple M = 〈W, R, D, V〉 such that:
— W is a non-empty set (of worlds);
— R is a binary (accessibility) relation on W;
— D is another non-empty set (of individuals);
— V is a function assigning
• an element V(a) ∈ D to every name a.
• a relation V(œ) ⊆ W × Dn to every n-ary predicate œ. (If œ is 0-ary, V(œ) is just the
proposition P expressed by œ, as before; otherwise V(œ) is the intension of œ..)
In such a model, an assignment µ is a function that maps every variable x to a value µ(x)∈D, and
two assignments µ and µ' are said to be x-alternatives iff they agree on every variable except
possibly for x [i.e., iff µ(y)=µ'(y) for every variable y distinct from x].
M M
5. 6=α ∀xA [µ] iff 6=α A [µ'] for every x-alternative of µ' of µ
M
— A wff A is true at α in M iff 6=α A [µ] for every assignment µ.
— The other semantic notions (validity in a model, validity in a class C of models, and logical
validity) are defined as before.
◆ REMARK 1
— The semantic condition on names corresponds to the idea that names are rigid designators.
◆ REMARK 2
— Concerning predicates, there is an expressiveness problem. Consider:
®∀xRx e.g., There is a world in which everybody is rich
— This formula refers to those existing at some world and says that they are all rich.
But do we mean to say that at some world everybody will be rich as they use the word
‘rich’, or as we use the word in the actual world?
2
— It could be argued that the above semantics is constrained to interpret wffs involving ‘R’ at a
world according to that world’s meaning of ‘R’, so it only gives us the first option. But this
is not quite right.
We certainly don’t mean to say that everybody could be rich insofar as ‘rich’ could have a
different meaning...
◆ LOGIC
— Modal logics adequate to this semantics can be axiomatized by supplementing the principles of
whichever propositional modal logic one chooses with the classical quantifier rules together
with the so-called Barcan Formula (Barcan 1946):
BF ∀xÉA → É∀xA
◆ REMARK 3
— BF captures one half of the “fixed domain” idea: it tells us that the domain of every possible
world is included in the domain of the given world. [Otherwise BF could be false, for even if
everything falling under the actual range of the quantifiers were A in every world, there might
be something else at some world that failed to be A at that world.]
— The other half of the “fixed domain” idea is captured by the converse of BF, which is valid
(and provable in quantified K even without BF).
This ensures that the domain of every possible world includes the domain of the given world.
[Otherewise BFc might be false, for even if everything in every world were A, there could be
something in the actual world that fails to exist (hence to be A) at some world.]
3
— If we have equality, this half of the “fixed domain” idea is also captured by the following valid
wff:
NecEx ∀x É ∃y(y = x)
◆ REMARK 4
— Note that we can read BF as asserting that if a necessity holds de re, then it also holds de dicto.
But we may as well read it as asserting that if a possibility holds de dicto, then it also holds de
re—for BF is equivalent to:
— Likewise, we can read BFc as asserting that if a necessity holds de dicto, then it holds de re.
But we may as well read it as asserting that if a possibility holds de re it holds de dicto:
— BFc and BFc® are more plausible. But they are also controversial. Consider:
• If necessarily everything exists, then everything necessarily exists.
• If there is something that might not have existed, then it is possible for there to be
something that doesn’t exist.
— In general, the problem with the “fixed domain” approach is that it seems a fundamental
feature of common ideas about modality (at least: alethic modality) that the existence of many
things is contingent, and that different objects exist in different possible worlds.
— However, one can avoid this objection by interpreting the quantifiers as ranging over all
possible objects and letting a designate predicate ‘E’ express existence.
For all existing x: A ⇒ ∀x(Ex → A)
For some existing x: A ⇒ ∃x(Ex ∧ A)
4
Modal Logic (G4424) Achille C. Varzi
◆ MODELS
A model is a five-tuple M = 〈W, R, D, V, Q〉 such that:
— W, R, D, V are as in Semantics #1
— Q is a function assigning a non-empty set Q(α) ⊆ D to each world α∈W (the domain of
quantification for α), with the proviso Q(α) ⊆ Q(β) whenever αRβ.
(This proviso ensures that BFc comes out valid, for recall: BFc is derivable in quantified K).
Assignments are defined as in #1.
◆ REMARKS 1
— The expressiveness problem persists.
— A (non-equivalent) alternative to clause 1 is:
M
6=α œt1 ...tn [µ] iff µ*(t1 ), ..., µ*(tn )〉∈Q(α) and 〈α, µ*(t1 ), ..., µ*(tn )〉∈V(œ)
— In this semantics, BF is not valid. For instance, the following instance of BF
∀xÉ œx → É∀x œx
is false in any world-relative model M such that:
W = {α, β},
D = {a, b}
Q(α) = {a}, Q(β) = {a, b}
V(œ) = {〈α, a〉, 〈β, a〉}
— However, if R is symmetrical then BF holds, since the proviso on Q implies that
Q(α) = Q(β) whenever αRβ.
1
Since symmetry corresponds to principle B, this means that BF is a theorem of KB. Proof:
1. ∀xÉA → ÉA QL axiom
2. ®∀xÉA → ®ÉA 1, RK®
3. ®ÉA → A B®
4. ®∀xÉA → A 2,3, PL
M M M
9. ®∀xÉA → ∀xA as in Lecture Notes 11
10. É®∀xÉA → É∀xA 9, RK
11. ∀xÉA → É®∀xÉA B
12. ∀xÉA → É∀xA 9,10, PL
This is disappointing, since symmetry is a plausible requrement on R (at least for alethic
modalities).
— Moreover, the validity of BFc, and more generally of the proviso that Q(α) ⊆ Q(β) whenever
αRβ, is still controversial. Can we give it up?
— We can give up BFc, but the resulting semantics does not validate classical logic. We need
supplement K with the axioms & rules of Free Logic. In particular, the QL axiom
∀xA → A [t/x]
[which was used in the proof of BFc.] must be weakened to:
∀y(∀xA → A [y/x])
or, if we have equality, to:
(∀xA ∧ ∃y(y = t)) → A [t/x]
— If we do so, then (as expected) BFc is valid in a model iff the inclusiveness condition
αRβ ⇒ Q(α) ⊆ Q(β)
is explicitly imposed upon R.
— If we also require R to be symmetric, then BF is valid too.
◆ Axioms
— If the language includes the identity predicate =, the standard identity axioms hold:
ID1 t=t
ID2 t = t' → (A → A [t'/t])
2
— As a consequence, we immediately get the following logical truths (provable regardless of
whether we rely on a Classical or Free Logic for the quantifiers):
NecId t = t' → É t = t'
NecDiv t ≠ t' → É t ≠ t'
◆ Identity or counterparthood?
— Both types of semantics assume that objects may exist in more than one world, and imply that
names are rigid designators.
— But one could object to this view on philosophical grounds. One could say that when we
counterfactualize about someone, say John, we don’t look at the ways John could have been
but, rather, at the way John’s counterparts are—his counterparts being those individuals that
somehow correspond to John in other worlds.
This means that a statement such as ‘John could have been a singer’ would be true, not if there
is a possible world in which John is a singer, but if there is a counterpart of John’s, in some
possible world, who is a singer.
— We can account for this intuition as follows (though a fuller account will be seen next week,
devoted to David Lewis’s Counterpart theory):
◆ MODELS
A model is a six-tuple M = 〈W, R, D, V, Q, C〉 such that:
— W, R, D are as in #1 and #2
— Q is as in #2, but with the proviso that Q(α) ∩ Q(β) = ∅ for all distinct α,β∈ W.
— C is a binary relation (of counterpart) on D.
— V is a function assigning
• a function V(a): W → D to every name a, with the proviso that V(a)(α)∈Q(α). (Intuitively,
V(a)(α) is the individual denoted by ‘a’ in α.)
• a relation V(œ) ⊆ W × Dn to every n-ary predicate œ (as before.)
3
Modal Logic (G4424) Achille C. Varzi
◆ Primitives:
Wx (x is a possible world)
Ixy (x is in possible world y)
Ax (x is actual)
Cxy (x is a counterpart of y)
◆ Axioms:
A1 Ixy → Wy
(Nothing is in anything except a world)
A2 Ixy ∧ Ixz → y=z
(Nothing is in two worlds)
A3 Cxy → ∃zIxz
(Whatever is a counterpart is in a world)
A4 Cxy → ∃zIyz
(Whatever has a counterpart is in a world)
A5 Ixy ∧ Izy ∧ Cxz → x=z
(Nothing is a counterpart of anything else in its world)
A6 Ixy → Cxx
(Anything in a world is a counterpart of itself)
A7 ∃x(Wx ∧ ∀y(Iyx ↔ Ay))
(Some world contains all and only actual things)
A8 ∃xAx
(Something is actual)
2. Remarks
1
— Ad A2: Worlds do not overlap; thus, possible individuals in different worlds are never identical
(cross-world identity is replaced by the counterpart relation). However, the possible individuals
are not all the individuals: cross-world mereological fusions of possible individuals are
individuals too, though not possible individuals: there is no way for the whole of it to be actual.
— Ad A3–A4: Only possible individuals are (and have) counterparts. My counterparts are
individuals I would have been, had the world been otherwise.
— Ad A5–A6: The counterpart relation is essentially a cross-world relation, with the only
exception that everything qualifies as a counterpart of itself.
— Ad A7–A8: There exists a unique actual world. Its description can safely be used:
@ = ιx∀y(Iyx ↔ Ay)
◆ Translation:
T1 Α ⇒ A@
where Aw (A holds in a world w) is defined recursively as follows:
T2a Aw = A , if A is atomic
T2b (¬A)w = ¬Aw
T2c (A ∧ B)w = Aw ∧ Bw
T2d (∀xA)w = ∀x(Ixw → Aw)
T2f0 ( A)w = ∀z(Wz → Az)
(A holds in every world z)
2
T2f1 ( A x)w = ∀z∀y(Wz ∧ Iyz ∧ Cyx → Azy)
(A holds of every counterpart y of x in every world z)
T2fn ( Ax1 …xn )w = ∀z∀y1 …∀yn (Wz ∧ Iy1 z ∧ Cy1 x1 ∧ … ∧ Iyn z ∧ Cyn xn → Azy1 …yn )
◆ Examples:
E1 ∀xFx
⇒ ∀x(Ix@ → Fx)
(Everything actual is an F)
E2 ∃xFx
⇒ ∃w(Ww ∧ ∃x(Ixw ∧ Fx))
(Some possible world contains an F)
E3 Fx
⇒ ∀z∀y(Wz ∧ Iyz ∧ Cyx → Fy)
(Every counterpart of x, in any world, is an F)
E4 ∀x(Fx → Fx)
⇒ ∀x(Ix@ →∀z∀y(Wz ∧ Iyz ∧ Cyx → Fx))
(If anything is a counterpart of an actual F, then it is an F)
E5 Fx
⇒ ∀z1 ∀y1 (Wz1 ∧ Iy1 z1 ∧ Cy1 x → ∃z2 ∃y2 (Wz2 ∧ Iy2 z2 ∧ Cy2 y1 ∧ Fy2 )
(Every counterpart of x has a counterpart that is an F)
◆ Critical principles:
B A→ A
Not a theorem (for A open) unless R1 (symmetry of C) is assumed
4 A→ A
Not a theorem (for A open) unless R2 (transitivity of C) is assumed
BF ∀x Ax → ∀xAx
Not a theorem unless R5 is assumed.
BF' ∃x Ax → ∃xAx
Not a theorem unless R6 is assumed.
BFc ∀xAx → ∀x Ax
A theorem.
BFc' ∃xAx → ∃x Ax
Not a theorem (obviously).
= x=y → (x=y)
Not a theorem unless R3 is assumed
≠ x≠y → (x≠y)
Not a theorem unless R4 is assumed.