Sequent Calculi for Normal Update Logics Katsuhiko Sano 1 Minghui Ma - - PowerPoint PPT Presentation

sequent calculi for normal update logics
SMART_READER_LITE
LIVE PREVIEW

Sequent Calculi for Normal Update Logics Katsuhiko Sano 1 Minghui Ma - - PowerPoint PPT Presentation

Sequent Calculi for Normal Update Logics Katsuhiko Sano 1 Minghui Ma 2 1 Graduate School of Letters, Hokkaido University, Japan 2 Institute of Logic and Coginition, Sun Yat-Sen University, China. ICLA 2019 @ IIT Delhi, 4th March 2019 Outline


slide-1
SLIDE 1

Sequent Calculi for Normal Update Logics

Katsuhiko Sano1 Minghui Ma2

1Graduate School of Letters, Hokkaido University, Japan 2Institute of Logic and Coginition, Sun Yat-Sen University, China.

ICLA 2019 @ IIT Delhi, 4th March 2019

slide-2
SLIDE 2

Outline

1

What are Normal Update Logics?

2

Background on Sequent Calculi for Modal Logics

3

Sequent Calculi for Normal Update Logics

slide-3
SLIDE 3

Kripke Semantics of Modal Logic

ψ: “It is necessary that ψ.” ϕ ::= p | ¬ϕ | ϕ → ϕ | ϕ. Let M = (W, R, V) where R ⊆ W × W. M, w | = ψ ⇐ ⇒ For all v (wRv implies M, v | = ψ). ψ is Gψ (ψ will be always the case) for Tense Logic

slide-4
SLIDE 4

Kripke Semantics of Tense Logic

ψ: “It was the case that ψ. ” (Pψ) ϕ ::= p | ¬ϕ | ϕ → ϕ | ϕ | ϕ. Let M = (W, R, V) where R ⊆ W × W. M, w | = ψ ⇐ ⇒ For some v (vRw and M, v | = ψ). | = ϕ → ψ ⇐ ⇒ | = ϕ → ψ.

slide-5
SLIDE 5

Kripke Sem. of Conditional Logic (Chellas 1975)

[ϕ]ψ: “If ϕ then (normally) ψ.” “The current belief base is updated by ϕ, ψ follows.” ϕ ::= p | ¬ϕ | ϕ → ϕ | [ϕ]ϕ. Let M = (W, (RX)X⊆W, V) where RX ⊆ W × W. M, w | = [ϕ]ψ ⇐ ⇒ For all v (wRϕv implies M, v | = ψ), where ϕ := { x ∈ W | M, x | = ϕ }. wRXv: v is one of the most “X-similar” states from w.

slide-6
SLIDE 6

Kripke Sem. of Normal Update Logic

ϕ−ψ: “ψ has been updated by ϕ.” (cf. Herzig (1998)) ϕ ::= p | ¬ϕ | ϕ → ϕ | [ϕ]ϕ | ϕ−ϕ. Let M = (W, (RX)X⊆W, V) where RX ⊆ W × W. M, w | = ϕ−ψ ⇐ ⇒ For some v (vRϕw and M, v | = ψ). where ϕ := { x ∈ W | M, x | = ϕ }. wRXv: v is one of the most “X-similar” states from w.

slide-7
SLIDE 7

Motivation for Normal Update Logic

ϕ−ψ: “ψ has been updated by ϕ.” (cf. Herzig (1998)) Let p be an input, q a current belief base, r a resulting belief base. p = “I come to New Delhi” q = “I suffer from a heavy jet-lag in Japan,” r = “My jet-lag becomes lighter.” p−q ⊢ r ⇐ ⇒ q ⊢ [p]r

Herzig (1998), p.193

we neither consider updates to be more basic than conditionals nor the contrary, and shall rather take the equivalence to be basic.

slide-8
SLIDE 8

HCK for Conditional Logic: Chellas 1975

(Taut) All instances of propositional tautologies (MP) Modus Ponens (K) [ϕ](ψ → θ) → ([ϕ]ψ → [ϕ]θ) (Nec) From ψ we may infer [ϕ]ψ. (EQCA) From ϕ1 ↔ ϕ2, we may infer [ϕ1]ψ ↔ [ϕ2]ψ.

slide-9
SLIDE 9

HUCK for Normal Update Logic: Herzig 1998

To Hilbert system HCK for Conditional Logic, we add: (Conv1) ψ → [ϕ]ϕ−ψ. (Conv2) ϕ−[ϕ]ψ → ψ. (Mon ·−) From ψ1 → ψ2 we may infer ϕ−ψ1 → ϕ−ψ2. (EQUA) From ϕ1 ↔ ϕ2, we may infer ϕ−

1 ψ ↔ ϕ− 2 ψ.

⊢HUCK ϕ−ψ → θ ⇐ ⇒ ⊢HUCK ψ → [ϕ]θ

Semantic Completeness of HUCK (Herzig 1998)

ϕ is provable in HUCK iff ϕ is valid in all models. (∵) By Canonical Model Construction.

slide-10
SLIDE 10

Background & Contribution of This Talk

Main Question of This Talk

Does HUCK enjoy the decidability or the finite model property (FMP)? This talk provides sequent calculus GUCK; shows that it is equipollent with HUCK; establishes the subformula property and FMP of GUCK to obtain the decidability.

slide-11
SLIDE 11

What are Sequents?

a sequent = a pair of finite sets of formulas ϕ1, . . . , ϕm ⇒ ψ1, . . . , ψn. “If all ϕis hold, then some ψj holds. ” (ϕ1 ∧ · · · ∧ ϕm) → (ψ1 ∨ · · · ∨ ψn)

slide-12
SLIDE 12

Sequent Calculus GK for Modal Logic K

Axioms: ϕ ⇒ ϕ Weakening rules Logical Rules: ϕ, Γ ⇒ ∆ Γ ⇒ ∆, ¬ϕ (⇒ ¬) Γ ⇒ ∆, ϕ ¬ϕ, Γ ⇒ ∆ (¬ ⇒) ϕ, Γ ⇒ ∆, ψ Γ ⇒ ∆, ϕ → ψ (⇒→) Γ ⇒ ∆, ϕ ψ, Σ ⇒ Π ϕ → ψ, Γ, Σ ⇒ ∆, Π (→⇒) Cut Rule: Γ ⇒ ∆, ϕ ϕ, Π ⇒ Σ Γ, Π ⇒ ∆, Σ (Cut) Modal Rule: ψ1, . . . , ψn ⇒ ϕ ψ1, . . . , ψn ⇒ ϕ () (n 0)

slide-13
SLIDE 13

Cut Elimination of GK

If a sequent is provable in GK then it is provable in GK w/o (Cut).

(Cor.) Subformula Property of GK

If Γ ⇒ ∆ is provable in GK then it is provable in GK by a derivation which consists of subformulas of Γ, ∆ alone.

slide-14
SLIDE 14

Sequent Calculus GKt for Tense Logic

To get GKt, we replace the rule () in GK w/: Θ, Γ ⇒ ϕ Θ, Γ ⇒ ϕ (Kt) ϕ ⇒ Σ, Θ ϕ ⇒ Σ, Θ (Kt) (Nishimura 1980) (Cut) is indispensable in GKt as: ¬p ⇒ ¬p ¬p ⇒ ¬p (Kt) p ⇒ p ¬p, p ⇒ (¬ ⇒) p, ¬p ⇒ (Cut)

Subformula Property of GKt (Takano 1992)

If Γ ⇒ ∆ is provable in GKt then it is provable in GKt by a derivation which consists of subformulas of Γ, ∆ alone.

slide-15
SLIDE 15

Takano’s Methods for Subformual Property

Syntactic Method: Very much like cut-elimination.

Takano (1992) Subformula property as a substitute for cut-elimination in modal propositional logics, Math Jpn, 37(6),1129-1145.

Semantic Method: Show that system w/ analytic cut is semantically complete.

Takano (2018) A semantical analysis of cut-free calculi for modal logics, Rep. Math. Logic, 54, 43-65.

slide-16
SLIDE 16
  • Seq. Calc. GCK for Conditional Logic

To get GCK, add the following to the Boolean part of GK: ϕ0 ⇔ · · · ⇔ ϕn ψ1, . . . , ψn ⇒ ψ0 [ϕ1]ψ1, . . . , [ϕn]ψn ⇒ [ϕ0]ψ0 ([·]) (Pattinson et al. 2011)

Cut Elimination of GCK (Pattinson et al. 2011)

If a sequent is provable in GCK then it is provable in GCK w/o (Cut).

slide-17
SLIDE 17
  • Seq. Calc. GUCK for Normal Update Logic

To the Boolean part of GK, we add the following two: ϕ0 ⇔ · · · ⇔ ϕn ϕ−

1 θ1, . . . , ϕ− n θn, ψ1, . . . , ψn ⇒ ψ0

θ1, . . . , θn, [ϕ1]ψ1, . . . , [ϕn]ψn ⇒ [ϕ0]ψ0 ([·]UCK) ϕ0 ⇔ · · · ⇔ ϕn ψ0 ⇒ ψ1, . . . , ψn, [ϕ1]θ1, . . . , [ϕn]θn ϕ−

0 ψ0 ⇒ ϕ− 1 ψ1, . . . , ϕ− n ψn, θ1, . . . , θn

(·−UCK)

slide-18
SLIDE 18

Equipollence Result

ϕ is provable in HUCK iff ⇒ ϕ is provable in GUCK. (Cut) is indepensable in GUCK: [q]¬p ⇒ [q]¬p q−[q]¬p ⇒ ¬p (·−UCK) p ⇒ p ¬p, p ⇒ (¬ ⇒) p, q−[q]¬p ⇒ (Cut)

slide-19
SLIDE 19

Γ ⇒ ∆ is Ξ-provable in GUCK if there is a derivation D of the sequent such that D consists of formulas from Ξ.

Main Result of This Talk

TFAE:

1

Γ ⇒ ∆ is Sub(Γ, ∆)-provable in GUCK.

2

Γ ⇒ ∆ is provable in GUCK.

3

Γ → ∆ is valid in all finite models. ∵ (1) ⇒ (2) & (2) ⇒ (3) are easy. We focus on (3) ⇒ (1) below.

Corollary

GUCK enjoys the subformula property and FMP hence

  • decidability. Therefore, HUCK is also decidable.
slide-20
SLIDE 20

Proof Outline of (3) ⇒ (1)

We prove the contrapositive implication.

1

Suppose: Γ ⇒ ∆ is not Sub(Γ, ∆)-provable in GUCK.

2

Put Ξ := Sub(Γ, ∆) (finite!).

3

Extend Γ ⇒ ∆ to a Ξ-complete Γ+ ⇒ ∆+, where “Ξ-complete” means:

Γ+ ∪ ∆+ = Ξ. Γ+ ⇒ ∆+ is still not Ξ-provable in GUCK.

4

Define MΞ = (W Ξ, (RΞ

X)X⊆W Ξ, V Ξ) as:

W Ξ = all Ξ-complete sequents (finite!). RΞ

X is defined via ([·]UCK) and (·−UCK).

Π ⇒ Σ ∈ V Ξ(p) iff p ∈ Π.

5

Γ → ∆ is falsified in the finite MΞ.

slide-21
SLIDE 21

Main Result of This Talk

TFAE:

1

Γ ⇒ ∆ is Sub(Γ, ∆)-provable in GUCK.

2

Γ ⇒ ∆ is provable in GUCK.

3

Γ → ∆ is valid in all finite models.

Corollary

GUCK enjoys the subformula property and FMP hence

  • decidability. Therefore, HUCK is also decidable.
slide-22
SLIDE 22

Further Direction

The paper contains the results on HUCK extended w/: (CID) [ϕ]ϕ and/or (CMP) [ϕ]ψ → (ϕ → ψ). Further extension, say w/ (CLEM) [ϕ]ψ ∨ [ϕ]¬ψ. Syntactic proof of the subformula property of GUCK? Craig Interpolation Theorem for GUCK?

slide-23
SLIDE 23

Sequent Calculus GS5 for Modal Logic S5

To get GS5, we replace the rule () in GK w/: Γ ⇒ ∆, ψ Γ ⇒ ∆, ψ (⇒ S5) ϕ, Γ ⇒ ∆ ϕ, Γ ⇒ ∆ ( ⇒) (Cut) is indispensable in GS5 as: ¬p, p ⇒ ¬p, p ⇒ (S5 ⇒) p ⇒ ¬¬p (⇒ ¬) ⇒ ¬p, ¬¬p ⇒ ¬p, ¬¬p (⇒ S5) ¬¬p ⇒ ¬¬p (¬ ⇒) p ⇒ ¬¬p (Cut)

Subformula Property of GS5 (Takano 1992)

If Γ ⇒ ∆ is provable in GS5 then it is provable in GS5 by a derivation which consists of subformulas of Γ, ∆ alone.