The Logic of Conditional Beliefs: Neighbourhood Semantics and - - PowerPoint PPT Presentation

the logic of conditional beliefs neighbourhood semantics
SMART_READER_LITE
LIVE PREVIEW

The Logic of Conditional Beliefs: Neighbourhood Semantics and - - PowerPoint PPT Presentation

The Logic of Conditional Beliefs: Neighbourhood Semantics and Sequent Calculus Marianna Girlando, Sara Negri, Nicola Olivetti, Vincent Risch Aix Marseille Universit e, Laboratoire des Sciences de lInformation et des Syst` emes; University


slide-1
SLIDE 1

The Logic of Conditional Beliefs: Neighbourhood Semantics and Sequent Calculus

Marianna Girlando, Sara Negri, Nicola Olivetti, Vincent Risch

Aix Marseille Universit´ e, Laboratoire des Sciences de l’Information et des Syst` emes; University of Helsinki, Department of Philosophy

Advances in Modal Logics Budapest, August 30 - September 2, 2016

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 1 / 37

slide-2
SLIDE 2

Outline

(1) The logic CDL (2) Semantics (3) Labelled Sequent Calculus (4) Main results: Soundness, Termination and Completeness (5) Conclusions

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 2 / 37

slide-3
SLIDE 3

Outline

(1) The logic CDL (2) Semantics (3) Labelled Sequent Calculus (4) Main results: Soundness, Termination and Completeness (5) Conclusions

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 3 / 37

slide-4
SLIDE 4

The Logic of Conditional Beliefs (CDL)

The Logic of Conditional Beliefs Multi-agent modal epistemic logic, featuring the conditional belief operator: Beli(B|A), “agent i believes B having learnt A” Three-wise-men puzzle

  • Agent a believes that she is wearing a white hat: BelaWa
  • Agent a learns that agent b knows the colour of the hat that b herself is

wearing, and changes her beliefs: she is now convinced that she is wearing a black hat: Bela(Ba|KbWb ∨ KbBb) References Baltag and Smets (2006); Baltag and Smets (2008); Board (2004); Pacuit (2013).

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 4 / 37

slide-5
SLIDE 5

The Logic of Conditional Beliefs (CDL)

Language of CDL A := P | ⊥ | ¬A | A ∧ A | A ∨ A | A ⊃ A | Beli(A|A) Epistemic operators

  • Conditional belief (primitive): Beli(C|B), “agent i believes C, given B”
  • Unconditional belief (defined): BeliB =df Beli(B|⊤), “agent i believes B”
  • Knowledge (defined): KiB =df Beli(⊥|¬B), “agent i knows B”

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 5 / 37

slide-6
SLIDE 6

Axiomatic presentation of CDL [Board, 2004]

Inference rules (1) If ⊢ B, then ⊢ Beli(B|A) (epistemization rule) (2) If ⊢ A ⊃⊂ B, then ⊢ Beli(C|A) ⊃⊂ Beli(C|B) (rule of logical equivalence) Axioms Any axiomatization of the classical propositional calculus, plus: (3) (Beli(B|A) ∧ Beli(B ⊃ C|A)) ⊃ Beli(C|A) (distribution axiom) (4) Beli(A|A) (success axiom) (5) Beli(B|A) ⊃⊂ (Beli(C|A ∧ B) ⊃ Beli(C|A)) (minimal change principle 1) (6) ¬Beli(¬B|A) ⊃ (Beli(C|A ∧ B) ⊃⊂ Beli(B ⊃ C|A)) (minimal change principle 2) (7) Beli(B|A) ⊃ Beli(Beli(B|A)|C) (positive introspection) (8) ¬Beli(B|A) ⊃ Beli(¬Beli(B|A)|C) (negative introspection) (9) A ⊃ ¬Beli(⊥|A) (consistency axiom) The axiomatization is related to the AGM postulates of belief revision.

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 6 / 37

slide-7
SLIDE 7

Outline

(1) The logic CDL (2) Semantics (3) Labelled Sequent Calculus (4) Main results: Soundness, Termination and Completeness (5) Conclusions

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 7 / 37

slide-8
SLIDE 8

Epistemic Plausibility Models for CDL

Epistemic plausibility models [Board, 2004; Baltag and Smets, 2008; Pacuit, 2013] Let A be a set of agents; an epistemic plausibility model (EPM) has the form M = W, {∼i}i∈A, {i}i∈A, where

  • W is a non-empty set of elements called “worlds”;
  • for each i ∈ A, ∼i is an equivalence relation over W;
  • for each i ∈ A, i is a well-founded pre-order over W;
  • : Atm → P(W) is the evaluation for atomic formulas.

The relations ∼i and i satisfy the following properties:

  • Plausibility implies possibility: If w i v then w ∼i v
  • Local connectedness: If w ∼i v then w i v or v i w

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 8 / 37

slide-9
SLIDE 9

Epistemic Plausibility Models for CDL

Truth conditions for formulas in EPM

  • ¬A ≡ W − A
  • A ∧ B ≡ A ∩ B
  • A ∨ B ≡ A ∪ B
  • A ⊃ B ≡ (W − A) ∪ B
  • Beli(B|A) ≡ {x ∈ W | Mini([x]∼i ∩ A) ⊆ B}

where [x]∼i = {w | w ∼i x} and Mini(S) = {u ∈ S | ∀z ∈ S (u i z)} Theorem: Completeness of the axiomatization [Board, 2004] A formula A is a theorem of CDL if and only if it is valid in the class of epistemic plausibility models.

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 9 / 37

slide-10
SLIDE 10

Neighbourhood Models for CDL

Neighbourhood models

  • These models associate to each world a set of sets of worlds, used to

interpret modalities; they were originally proposed to give an interpretation of non-normal modal logics: Scott (1970), Montague (1970), Chellas (1980)...

  • Semantics of counterfactuals: Sphere models, Lewis (1973);
  • Semantics of belief revision: Grove (1988);
  • Studied recently also by Pacuit (2007); Marti and Pinosio (2013); Negri and

Olivetti (2015); Negri (2016).

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 10 / 37

slide-11
SLIDE 11

Neighbourhood Models for CDL

Multi-agent neighbourhood models Let A be a set of agents; a multi-agent neighbourhood model (NM) has the form M = W, {I }i∈A, where

  • W is a non empty set of elements called “worlds” ;
  • for each i ∈ A, Ii : W → P(P(W)) is the neighbourhood function, satisfying

the following properties:

Non-emptiness: ∀α ∈ Ii(x), α ∅ Nesting: ∀α, β ∈ Ii(x), α ⊆ β or β ⊆ α Total reflexivity: ∃α ∈ Ii(x) such that x ∈ α Local absoluteness: If α ∈ Ii(x) and y ∈ α then Ii(x) = Ii(y) Closure under intersection: If S ⊆ Ii(x) and S ∅ then S ∈ S (always holds in finite models)

  • : Atm → P(W) is the evaluation for atomic formulas.

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 11 / 37

slide-12
SLIDE 12

Neighbourhood Models for CDL

Forcing relation [Negri, 2016]

  • variables for worlds: x, y, z . . .
  • variables for neighbourhoods: α, β, γ . . .
  • “x forces A ”, for A formula:

x A iff x ∈ A

  • “α universally forces A ”:

α ∀ A iff ∀y ∈ α (y A)

  • “α existentially forces A ”:

α ∃ A iff ∃y ∈ α (y A) Truth conditions for formulas in NM

  • Truth conditions for propositional formulas are the ones defined for EPM

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 12 / 37

slide-13
SLIDE 13

Conditional Belief

Truth condition x Beli(B|A) iff ∀α ∈ Ii(x)(α ∩ A = ∅) or ∃β ∈ Ii(x)(β ∩ A ∅ and β ∩ A ⊆ B) iff ∀α ∈ Ii(x)(α ∀ ¬A)

  • r

∃β ∈ Ii(x)(β ∃ A and β ∀ A ⊃ B)

y z w x

x Ii(x) Ii B A β α

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 13 / 37

slide-14
SLIDE 14

Belief

Truth condition x BeliA iff ∃β ∈ Ii(x) (β ⊆ A) iff ∃β ∈ Ii(x) (β ∀ A)

y z x

x Ii(x) Ii A β α

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 14 / 37

slide-15
SLIDE 15

Knowledge

Truth condition x KiA iff ∀β ∈ Ii(x) (β ⊆ A) iff ∀β ∈ Ii(x) (β ∀ A)

y z x

x Ii(x) Ii A β α

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 15 / 37

slide-16
SLIDE 16

Equivalence Between Plausibility Models and Neighbourhood Models

Theorem: Equivalence between models A formula A is valid in the class of epistemic plausibility models if and only if it is valid in the class of multi-agent neighbourhood models. Proof. Generalization of the canonical “topological construction” considered by Pacuit (2013) and Marti and Pinosio (2013), and going back to Alexandroff (1937).

  • Corollary: Completeness of the axiomatization

A formula A is a theorem of CDL if and only if it is valid in the class of neighbourhood models.

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 16 / 37

slide-17
SLIDE 17

Outline

(1) The logic CDL (2) Semantics (3) Labelled Sequent Calculus (4) Main results: Soundness, Termination and Completeness (5) Conclusions

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 17 / 37

slide-18
SLIDE 18

A Labelled Sequent Calculus for CDL

Sequent calculus G3CDL G3CDL is a labelled sequent calculus which internalizes the neighbourhood semantics of CDL.

  • labels for worlds: x, y, z . . .
  • labels for neighbourhoods: a, b, c . . .
  • a ∃ A ≡ ∃x (x ∈ a and x A)
  • a ∀ A ≡ ∀x (x ∈ a implies x A)
  • x i B|A ≡ ∃c (c ∈ Ii(x) and c ∃ A and c ∀ A ⊃ B)
  • x Beli(B|A) ≡ ∀a ∈ Ii(x)(a ∀ ¬A) or x i B|A

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 18 / 37

slide-19
SLIDE 19

A Labelled Sequent Calculus for CDL

G3CDL Rules (1)

Initial sequents x : P, Γ ⇒ ∆, x : P Rules for local forcing x ∈ a, Γ ⇒ ∆, x : A Γ ⇒ ∆, a ∀ A

R∀ (x fresh)

x : A, x ∈ a, a ∀ A, Γ ⇒ ∆ x ∈ a, a ∀ A, Γ ⇒ ∆

L∀

x ∈ a, Γ ⇒ ∆, x : A, a ∃ A x ∈ a, Γ ⇒ ∆, a ∃ A

R∃

x ∈ a, x : A, Γ ⇒ ∆ a ∃ A, Γ ⇒ ∆

L∃ (x fresh)

Propositional rules: rules of G3K [Negri 2005]

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 19 / 37

slide-20
SLIDE 20

A Labelled Sequent Calculus for CDL

G3CDL Rules (2)

Rules for conditional belief a ∈ Ii(x), a ∃ A, Γ ⇒ ∆, x i B|A Γ ⇒ ∆, x : Beli(B|A)

RB (a fresh)

a ∈ Ii(x), x : Beli(B|A), Γ ⇒ ∆, a ∃ A x i B|A, a ∈ Ii(x), x : Beli(B|A), Γ ⇒ ∆ a ∈ Ii(x), x : Beli(B|A), Γ ⇒ ∆

LB

a ∈ Ii(x), Γ ⇒ ∆, x i B|A, a ∃ A a ∈ Ii(x), Γ ⇒ ∆, x i B|A, a ∀ A ⊃ B a ∈ Ii(x), Γ ⇒ ∆, x i B|A

RC

a ∈ Ii(x), a ∃ A, a ∀ A ⊃ B, Γ ⇒ ∆ x i B|A, Γ ⇒ ∆

LC(a fresh)

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 20 / 37

slide-21
SLIDE 21

A Labelled Sequent Calculus for CDL

G3CDL Rules (3)

Rules for inclusion a ⊆ a, Γ ⇒ ∆ Γ ⇒ ∆

Ref

c ⊆ a, c ⊆ b, b ⊆ a, Γ ⇒ ∆ c ⊆ b, b ⊆ a, Γ ⇒ ∆

Tr

x ∈ a, a ⊆ b, x ∈ b, Γ ⇒ ∆ x ∈ a, a ⊆ b, Γ ⇒ ∆

L⊆

Rules for semantic conditions a ⊆ b, a ∈ Ii(x), b ∈ Ii(x), Γ ⇒ ∆ b ⊆ a, a ∈ Ii(x), b ∈ Ii(x), Γ ⇒ ∆ a ∈ Ii(x), b ∈ Ii(x), Γ ⇒ ∆

S

x ∈ a, a ∈ Ii(x), Γ ⇒ ∆ Γ ⇒ ∆

T (a fresh)

a ∈ Ii(x), y ∈ a, b ∈ Ii(x), b ∈ Ii(y), Γ ⇒ ∆ a ∈ Ii(x), y ∈ a, b ∈ Ii(x), Γ ⇒ ∆

A1

a ∈ Ii(x), y ∈ a, a ∈ Ii(y), Γ ⇒ ∆ a ∈ Ii(x), y ∈ a, Γ ⇒ ∆

A2

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 21 / 37

slide-22
SLIDE 22

Derivation Example: Axiom (6) ¬Beli(¬B|A) ⊃ (Beli(B ⊃ C|A) ⊃ Beli(C|A ∧ B))

D                                        y : A · · · ⇒ . . . y : A y : B · · · ⇒ . . . y : B y : A, y : B, y ∈ b, c ∈ Ii(x), c ∃ A, b ∈ Ii(x) · · · ⇒ . . . y : A ∧ B (R∧) y : A, y : B, y ∈ b, c ∈ Ii(x), c ∃ A, b ∈ Ii(x) · · · ⇒ . . . b ∃ A ∧ B (R ∃) y ∈ b, c ∈ Ii(x), c ∃ A, b ∈ Ii(x) · · · ⇒ . . . b ∃ A ∧ B, y : A ⊃ ¬B (R ⊃, R¬) c ∈ Ii(x), c ∃ A, b ∈ Ii(x) · · · ⇒ . . . b ∃ A ∧ B, b ∀ A ⊃ ¬B (R ∀) c ∈ Ii(x), c ∃ A, b ∈ Ii(x) · · · ⇒ . . . b ∃ A ∧ B, x i ¬B|A (RC) b ∈ Ii(x), b ∃ A, b ∀ A ⊃ C, a ∃ A ∧ B · · · ⇒ . . . x : Beli(¬B|A), b ∃ A ∧ B (RB) E                          z : A · · · ⇒ . . . z : A z : c · · · ⇒ . . . z : C z : A ⊃ C, z : A, z : B, z ∈ b, b ∈ Ii(x), b ∃ A, b ∀ A ⊃ C, a ∃ A ∧ B, · · · ⇒ . . . z : C (L ⊃) z : A, z : B, z ∈ b, b ∈ Ii(x), b ∃ A, b ∀ A ⊃ C, a ∃ A ∧ B · · · ⇒ . . . z : C (L ∀) z ∈ b, b ∈ Ii(x), b ∃ A, b ∀ A ⊃ C, a ∃ A ∧ B · · · ⇒ . . . z : (A ∧ B) ⊃ C (R ⊃, L∧) b ∈ Ii(x), b ∃ A, b ∀ A ⊃ C, a ∃ A ∧ B · · · ⇒ . . . b ∀ (A ∧ B) ⊃ C (R ∀) D E b ∈ Ii(x), b ∃ A, b ∀ A ⊃ C, a ∈ Ii(x), a ∃ A ∧ B, x : Beli(C|A) ⇒ x : Beli(¬B|A), x i C|A ∧ B (RC) x i C|A, a ∈ Ii(x), a ∃ A ∧ B, x : Beli(C|A) ⇒ x : Beli(¬B|A), x i C|A ∧ B (LC) a ∈ Ii(x), a ∃ A ∧ B, x : Beli(C|A) ⇒ x : Beli(¬B|A), x i C|A ∧ B (LB) x : Beli(C|A) ⇒ x : Beli(¬B|A), x : Beli(C|A ∧ B) (RB) x : ¬(Beli(¬B|A)), x : Beli(C|A) ⇒ x : Beli(C|A ∧ B) (L¬)

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 22 / 37

slide-23
SLIDE 23

Structural Properties of G3CDL

Admissibility of Weakening The rules of left and right weakening are height-preserving admissible in G3CDL. Invertibility All the rules of G3CDL are height-preserving invertible. Admissibility of Contraction The rules of left and right contraction are height-preserving admissible in G3CDL. Admissibility of Cut Rule of Cut is admissible in G3CDL.

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 23 / 37

slide-24
SLIDE 24

Adding Knowlege and Belief

G3CDL Rules (4)

Rules for knowledge and belief a ∈ Ii(x), Γ ⇒ ∆, a ∀ A Γ ⇒ ∆, x : KiA

LK (a new)

a ∈ Ii(x), x : KiA, a ∀ A, Γ ⇒ ∆ a ∈ Ii(x), x : KiA, Γ ⇒ ∆

RK

a ∈ Ii(x), Γ ⇒ ∆, x : BeliA, a ∀ A a ∈ Ii(x), Γ ⇒ ∆, x : BeliA

LSB

a ∈ Ii(x), a ∀ A ⇒ ∆ x : BeliA, Γ ⇒ ∆

RSB (a new)

Admissibility of the rules The rules for knowledge and belief are admissible in G3CDL.

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 24 / 37

slide-25
SLIDE 25

Outline

(1) The logic CDL (2) Semantics (3) Labelled Sequent Calculus (4) Main results: Soundness, Termination and Completeness (5) Conclusions

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 25 / 37

slide-26
SLIDE 26

Main Results

Soundness If a sequent Γ ⇒ ∆ is derivable in G3CDL, then it is valid in the class of multi-agent neighbourhood models. Completeness If a formula A is valid in the class of multi-agent neighbourhood models, then it is derivable in G3CDL. Termination Adopting a suitable strategy, proof search for any sequent of the form ⇒ x0 : A always comes to an end after a finite number of steps. Finite model property If a formula A is satisfiable in the class of neighbourhood models, then it is satisfiable in the class of finite neighbourhood models.

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 26 / 37

slide-27
SLIDE 27

Proof sketch

Definition of saturated sequent Definition of a suitable proof search strategy Claim: Terminating derivation tree Each sequent that occurs as a leaf of a derivation tree built in accordance with the search strategy is either an initial sequent or a saturated sequent. Claim: Existence of a finite countermodel Let Γi ⇒ ∆i be a saturated sequent occurring as a leaf of a derivation branch. Then there exists a finite countermodel M to Γi ⇒ ∆i that satisfies all formulas in ↓ Γi and falsifies all formulas in ↓ ∆i (where ↓Γi =

j≤i Γj and ↓∆i = j≤i ∆j).

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 27 / 37

slide-28
SLIDE 28

Outline

(1) The logic CDL (2) Semantics (3) Labelled Sequent Calculus (4) Main results: Soundness, Termination and Completeness (5) Conclusions

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 28 / 37

slide-29
SLIDE 29

Conclusions

Results

  • A new simple neighbourhood semantics for CDL
  • A labelled sequent calculus based on it with good properties:
  • analyticity, cut-freeness
  • terminating proof-search
  • Constructive proof of the finite model property of CDL

Future Research

  • Provide an interpretation in NM of other epistemic operators defined in the

literature: safe belief, strong belief [Baltag and Smets, 2008];

  • Provide a direct proof of completeness of the axiomatization with respect to

the semantics defined in terms of neighbourhood models;

  • Long term goal: to obtain modular and uniform calculi covering all logics at

least as strong as CDL, including the family of Lewis’ logic of counterfactuals.

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 29 / 37

slide-30
SLIDE 30

Thank you !

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 30 / 37

slide-31
SLIDE 31

References

Alexandroff, P . (1937). Diskrete R¨

  • aume. Mat.Sb. (NS), 2(3):501–519.

Baltag, A. and Smets, S. (2006). Conditional doxastic models: A qualitative approach to dynamic belief

  • revision. Electronic Notes in Theoretical Computer Science, 165:5–21.

Baltag, A. and Smets, S. (2008). A qualitative theory of dynamic interactive belief revision. Logic and the foundations of game and decision theory (LOFT 7), 3:9–58. Board, O. (2004). Dynamic interactive epistemology. Games and Economic Behavior, 49(1):49–80. Chellas, B. F. (1980). Modal logic: an introduction, volume 316. Cambridge University Press. Grove, A. (1988). Two modellings for theory change. Journal of philosophical logic, 17(2):157–170. Lewis, D. K. (1973). Counterfactuals. Blackwell. Marti, J. and Pinosio, R. (2013). Topological semantics for conditionals. The Logica Yearbook. Montague, R. (1970). Pragmatics and intensional logic. Synthese, 22(1-2):68–94. Negri, S. (2016). Proof theory for non-normal modal logics: The neighbourhood formalism and basic

  • results. to appear,. http://www.helsinki.fi/˜negri/negri_ifcolog.pdf.

Negri, S. and Olivetti, N. (2015). A sequent calculus for preferential conditional logic based on neighbourhood semantics. In Automated Reasoning with Analytic Tableaux and Related Methods, pages 115–134. Springer. Pacuit, E. (2007). Neighborhood semantics for modal logic. Notes of a course on neighborhood structures for modal logic: http://ai.stanford.edu/ epacuit/classes/esslli/nbhdesslli.pdf. Pacuit, E. (2013). Dynamic epistemic logic I: Modeling knowledge and belief. Philosophy Compass, 8(9):798–814. Scott, D. (1970). Advice on modal logic. In Philosophical problems in logic, pages 143–173. Springer.

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 31 / 37

slide-32
SLIDE 32

Equivalence of Plausibility Models and Neighbourhood Models (1)

Theorem 1 if a formula A is valid in the class of multi-agent Neighbourhood Models then it is valid in the class of Epistemic Plausibility Models Proof. Let MP = W, {∼i}i∈A, {i}i∈A, [ ] be an P-model. Let u ∈ W define its downward closed set: ↓i u = {v ∈ W | v i u} We define the N-modelmodel MN = W, {I}i∈A, [ ], where for x ∈ W Ii(x) = {↓i u | u ∼i x}

  • Girlando, Negri, Olivetti, Risch

The Logic of Conditional Beliefs 32 / 37

slide-33
SLIDE 33

Equivalence of Plausibility Models and Neighbourhood Models (2)

Theorem 2 if a formula A is valid in the class of Epistemic Plausibility Models then it is valid in the class of multi-agent Neighbourhood Models Proof. Let MN = W, {I}i∈A, [ ] be a multi-agent N-model. We construct an P-model MP = W, {∼i}i∈A, {i}i∈A, [ ], by stipulating: x ∼i y iff ∃α ∈ Ii(x), y ∈ α x i y iff ∀α ∈ Ii(y), if y ∈ α then x ∈ α.

  • Corollary

A formula A is a theorem of CDL if and only if it is valid in the class of neighbourhood models.

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 33 / 37

slide-34
SLIDE 34

Adding Knowlege and Belief

Admissibility of LK in G3CDL KiA =df Beli(⊥|¬A) a ∈ Ii(x), Γ ⇒ ∆, a ∀ A Γ ⇒ ∆, x : KiA

LK (a new)

a ∈ Ii(x), a ∃ ¬A, Γ ⇒ ∆, x i ⊥|¬A, a ∃ ¬A a ∈ Ii(x), Γ ⇒ ∆, a ∀ A a ∈ Ii(x), a ∃ ¬A, Γ ⇒ ∆, x i ⊥|¬A, a ∀ A

Wk

a ∈ Ii(x), a ∃ ¬A, Γ ⇒ ∆, x i ⊥|¬A

RC

Γ ⇒ ∆, x : Beli(⊥|¬A)

RB

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 34 / 37

slide-35
SLIDE 35

Derivation Example: Axiom (9) A ⊃ ¬Beli(⊥|A)

D : x ∈ a, a ∈ Ii(x), x : A, x : Beli(⊥|A) ⇒ a ∃ A, x : A x ∈ a, a ∈ Ii(x), x : A, x : Beli(⊥|A) ⇒ a ∃ A

R∃

. . . . D y ∈ b, y : A, b ∈ Ii(x), b ∃ A, b ∀ A ⊃ ⊥, x ∈ a, a ∈ Ii(x), x : A, x : Beli(⊥|A) ⇒ y : A; y : ⊥, y ∈ b, y : A, b ∈ Ii(x), b ∃ A, b ∀ A ⊃ ⊥, x ∈ a, a ∈ Ii(x), x : A, x : Beli(⊥|A) ⇒ y : A ⊃ ⊥, y ∈ b, y : A, b ∈ Ii(x), b ∃ A, b ∀ A ⊃ ⊥, x ∈ a, a ∈ Ii(x), x : A, x : Beli(⊥|A) ⇒

L⊃

y ∈ b, y : A, b ∈ Ii(x), b ∃ A, b ∀ A ⊃ ⊥, x ∈ a, a ∈ Ii(x), x : A, x : Beli(⊥|A) ⇒

L∀

b ∈ Ii(x), b ∃ A, b ∀ A ⊃ ⊥, x ∈ a, a ∈ Ii(x), x : A, x : Beli(⊥|A) ⇒

L∃

x i ⊥|A, x ∈ a, a ∈ Ii(x), x : A, x : Beli(⊥|A) ⇒

LC

x ∈ a, a ∈ Ii(x), x : A, x : Beli(⊥|A) ⇒

LB

x : A, x : Beli(⊥|A) ⇒

T

x : A ⇒ x : ¬Beli(⊥|A)

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 35 / 37

slide-36
SLIDE 36

Main results

Saturated sequent Consider a derivation branch of the form Γ0 ⇒ ∆0, ..., Γk ⇒ ∆k, Γk+1 ⇒ ∆k+1, ... where Γ0 ⇒ ∆0 is the sequent ⇒ x0 : A, and ↓Γi =

j≤i Γj and ↓∆i = j≤i ∆j. For

each rule (R), we say that a sequent Γ ⇒ ∆ satisfies the saturation condition associated to (R) if the following hold: (R ∀) If a ∀ A is in ↓ ∆, then for some x there is x ∈ a in Γ and x : A in ↓ ∆; (L ∀) If x ∈ a and a ∀ A are in Γ, then x : A is in Γ; (RB) If x : Beli(B|A) is in ↓ ∆, then for some i ∈ A and for some a, a ∈ Ii(x) is in Γ, a ∃ A is in ↓ Γ and x i B|A is in ↓ ∆; (LB) If a ∈ Ii(x) and x : Beli(B|A) are in Γ, then either a ∃ A is in ↓ ∆ or x i B|A is in ↓ Γ; (T) For all x occurring in ↓ Γ∪ ↓ ∆, for all i ∈ A there is an a such that a ∈ Ii(x) and x ∈ a are in Γ; (S) If a ∈ Ii(x) and b ∈ Ii(x) are in Γ, then a ⊆ b or b ⊆ a are in Γ; ... A sequent Γ ⇒ ∆ is saturated if (Init) There is no x : P in Γ ∩ ∆; (L⊥) There is no x : ⊥ in Γ; Γ ⇒ ∆ satisfies all saturation conditions listed above.

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 36 / 37

slide-37
SLIDE 37

Main results

Proof search strategy When constructing root-first a derivation tree for a sequent ⇒ x0 : A, apply the following strategy: (1) No rule can be applied to an initial sequent; (2) If k(x) < k(y) all rules applicable to x are applied before any rule applicable to y. (3) Rule (T) is applied as the first one to each world label x. (4) Rules which do not introduce a new label (static rules) are applied before the rules which do introduce new labels (dynamic rules), with the exception of (T), as in (iii); (5) Rule (RB) is applied before rule (LC); (6) A rule (R) cannot be applied to a sequent Γi ⇒ ∆i if ↓ Γi and / or ↓ ∆i satisfy the saturation condition associated to (R).

Girlando, Negri, Olivetti, Risch The Logic of Conditional Beliefs 37 / 37