Uniform Interpolation and Proof Systems Rosalie Iemhoff Utrecht - - PowerPoint PPT Presentation

uniform interpolation and proof systems
SMART_READER_LITE
LIVE PREVIEW

Uniform Interpolation and Proof Systems Rosalie Iemhoff Utrecht - - PowerPoint PPT Presentation

Uniform Interpolation and Proof Systems Rosalie Iemhoff Utrecht University Workshop on Admissible Rules and Unification II Les Diablerets, February 2, 2015 1 / 12 Proof systems An old question: When does a logic have a decent proof system? 2


slide-1
SLIDE 1

Uniform Interpolation and Proof Systems

Rosalie Iemhoff Utrecht University Workshop on Admissible Rules and Unification II Les Diablerets, February 2, 2015

1 / 12

slide-2
SLIDE 2

Proof systems An old question: When does a logic have a decent proof system?

2 / 12

slide-3
SLIDE 3

Proof systems An old question: When does a logic have a decent proof system? Another old question: When does a logic have a sequent calculus?

2 / 12

slide-4
SLIDE 4

Proof systems An old question: When does a logic have a decent proof system? Another old question: When does a logic have a sequent calculus? Answers: Many positive instances. Less negative ones.

2 / 12

slide-5
SLIDE 5

Proof systems An old question: When does a logic have a decent proof system? Another old question: When does a logic have a sequent calculus? Answers: Many positive instances. Less negative ones. Related work:

2 / 12

slide-6
SLIDE 6

Proof systems An old question: When does a logic have a decent proof system? Another old question: When does a logic have a sequent calculus? Answers: Many positive instances. Less negative ones. Related work: (Negri) Fix a labelled sequent calculus and determine which axioms, when added, preserve cut-elimination.

2 / 12

slide-7
SLIDE 7

Proof systems An old question: When does a logic have a decent proof system? Another old question: When does a logic have a sequent calculus? Answers: Many positive instances. Less negative ones. Related work: (Negri) Fix a labelled sequent calculus and determine which axioms, when added, preserve cut-elimination. (Ciabattoni, Galatos, Terui) Fix a sequent calculus and determine which axioms

  • r structural rules, when added, preserve cut-elimination.

2 / 12

slide-8
SLIDE 8

Proof systems An old question: When does a logic have a decent proof system? Another old question: When does a logic have a sequent calculus? Answers: Many positive instances. Less negative ones. Related work: (Negri) Fix a labelled sequent calculus and determine which axioms, when added, preserve cut-elimination. (Ciabattoni, Galatos, Terui) Fix a sequent calculus and determine which axioms

  • r structural rules, when added, preserve cut-elimination.

Aim: Formulate properties that, when violated by a logic, imply that the logic does not have a sequent calculus of a certain form.

2 / 12

slide-9
SLIDE 9

Uniform interpolation Dfn A logic L has interpolation if whenever ⊢ ϕ → ψ there is a χ in the common language L(ϕ) ∩ L(ψ) such that ⊢ ϕ → χ and ⊢ χ → ψ.

3 / 12

slide-10
SLIDE 10

Uniform interpolation Dfn A logic L has interpolation if whenever ⊢ ϕ → ψ there is a χ in the common language L(ϕ) ∩ L(ψ) such that ⊢ ϕ → χ and ⊢ χ → ψ. Dfn A propositional (modal) logic has uniform interpolation if the interpolant depends only on the premiss or the conclusion: For all ϕ there are formulas ∃pϕ and ∀pϕ not containing p such that for all ψ not containing p: ⊢ ψ → ϕ ⇔ ⊢ ψ → ∀pϕ ⊢ ϕ → ψ ⇔ ⊢ ∃pϕ → ψ.

3 / 12

slide-11
SLIDE 11

Uniform interpolation Dfn A logic L has interpolation if whenever ⊢ ϕ → ψ there is a χ in the common language L(ϕ) ∩ L(ψ) such that ⊢ ϕ → χ and ⊢ χ → ψ. Dfn A propositional (modal) logic has uniform interpolation if the interpolant depends only on the premiss or the conclusion: For all ϕ there are formulas ∃pϕ and ∀pϕ not containing p such that for all ψ not containing p: ⊢ ψ → ϕ ⇔ ⊢ ψ → ∀pϕ ⊢ ϕ → ψ ⇔ ⊢ ∃pϕ → ψ. Algebraic view (next talk).

3 / 12

slide-12
SLIDE 12

Uniform interpolation Dfn A logic L has interpolation if whenever ⊢ ϕ → ψ there is a χ in the common language L(ϕ) ∩ L(ψ) such that ⊢ ϕ → χ and ⊢ χ → ψ. Dfn A propositional (modal) logic has uniform interpolation if the interpolant depends only on the premiss or the conclusion: For all ϕ there are formulas ∃pϕ and ∀pϕ not containing p such that for all ψ not containing p: ⊢ ψ → ϕ ⇔ ⊢ ψ → ∀pϕ ⊢ ϕ → ψ ⇔ ⊢ ∃pϕ → ψ. Algebraic view (next talk). Note A locally tabular logic that has interpolation, has uniform interpolation. ∃pϕ(p, ¯ q) = {ψ(¯ q) | ⊢ ϕ(p, ¯ q) → ψ(¯ q)} ∀pϕ(p, ¯ q) = {ψ(¯ q) | ⊢ ψ(¯ q) → ϕ(p, ¯ q)}

3 / 12

slide-13
SLIDE 13

Modal and intermediate logics Thm (Pitts ’92) IPC has uniform interpolation.

4 / 12

slide-14
SLIDE 14

Modal and intermediate logics Thm (Pitts ’92) IPC has uniform interpolation. Thm (Shavrukov ’94) GL has uniform interpolation.

4 / 12

slide-15
SLIDE 15

Modal and intermediate logics Thm (Pitts ’92) IPC has uniform interpolation. Thm (Shavrukov ’94) GL has uniform interpolation. Thm (Ghilardi & Zawadowski ’95) K has uniform interpolation. S4 does not.

4 / 12

slide-16
SLIDE 16

Modal and intermediate logics Thm (Pitts ’92) IPC has uniform interpolation. Thm (Shavrukov ’94) GL has uniform interpolation. Thm (Ghilardi & Zawadowski ’95) K has uniform interpolation. S4 does not. Thm (Bilkova ’06) KT has uniform interpolation. K4 does not.

4 / 12

slide-17
SLIDE 17

Modal and intermediate logics Thm (Pitts ’92) IPC has uniform interpolation. Thm (Shavrukov ’94) GL has uniform interpolation. Thm (Ghilardi & Zawadowski ’95) K has uniform interpolation. S4 does not. Thm (Bilkova ’06) KT has uniform interpolation. K4 does not. Thm (Maxsimova ’77, Ghilardi & Zawadowski ’02) There are exactly seven intermediate logics with (uniform) interpolation: IPC, Sm, GSc, LC, KC, Bd2, CPC.

4 / 12

slide-18
SLIDE 18

Modal and intermediate logics Thm (Pitts ’92) IPC has uniform interpolation. Thm (Shavrukov ’94) GL has uniform interpolation. Thm (Ghilardi & Zawadowski ’95) K has uniform interpolation. S4 does not. Thm (Bilkova ’06) KT has uniform interpolation. K4 does not. Thm (Maxsimova ’77, Ghilardi & Zawadowski ’02) There are exactly seven intermediate logics with (uniform) interpolation: IPC, Sm, GSc, LC, KC, Bd2, CPC. Pitts uses Dyckhoff’s ’92 sequent calculus for IPC.

4 / 12

slide-19
SLIDE 19

Modularity Aim: If a modal or intermediate logic has such an such a sequent calculus, then it has uniform interpolation.

5 / 12

slide-20
SLIDE 20

Modularity Aim: If a modal or intermediate logic has such an such a sequent calculus, then it has uniform interpolation. Therefore no modal or intermediate logic without uniform interpolation has such an such a calculus.

5 / 12

slide-21
SLIDE 21

Modularity Aim: If a modal or intermediate logic has such an such a sequent calculus, then it has uniform interpolation. Therefore no modal or intermediate logic without uniform interpolation has such an such a calculus. Modularity: The possibility to determine whether the addition of a new rule will preserve uniform interpolation.

5 / 12

slide-22
SLIDE 22

Focussed rules Dfn Multiplication of sequents: (Γ ⇒ ∆) · (Π ⇒ Σ) ≡ (Γ, Π ⇒ ∆, Σ).

6 / 12

slide-23
SLIDE 23

Focussed rules Dfn Multiplication of sequents: (Γ ⇒ ∆) · (Π ⇒ Σ) ≡ (Γ, Π ⇒ ∆, Σ). Dfn A rule is focussed if it is of the form S · S1 . . . S · Sn S · S0 where S, Si are sequents and S0 contains exactly one formula.

6 / 12

slide-24
SLIDE 24

Focussed rules Dfn Multiplication of sequents: (Γ ⇒ ∆) · (Π ⇒ Σ) ≡ (Γ, Π ⇒ ∆, Σ). Dfn A rule is focussed if it is of the form S · S1 . . . S · Sn S · S0 where S, Si are sequents and S0 contains exactly one formula. Ex The following rules are focussed. Γ ⇒ A, ∆ Γ ⇒ B, ∆ Γ ⇒ A ∧ B, ∆

6 / 12

slide-25
SLIDE 25

Focussed rules Dfn Multiplication of sequents: (Γ ⇒ ∆) · (Π ⇒ Σ) ≡ (Γ, Π ⇒ ∆, Σ). Dfn A rule is focussed if it is of the form S · S1 . . . S · Sn S · S0 where S, Si are sequents and S0 contains exactly one formula. Ex The following rules are focussed. Γ ⇒ A, ∆ Γ ⇒ B, ∆ Γ ⇒ A ∧ B, ∆ Γ, B → C ⇒ A → B Γ, C ⇒ D Γ, (A → B) → C ⇒ D

6 / 12

slide-26
SLIDE 26

Focussed rules Dfn Multiplication of sequents: (Γ ⇒ ∆) · (Π ⇒ Σ) ≡ (Γ, Π ⇒ ∆, Σ). Dfn A rule is focussed if it is of the form S · S1 . . . S · Sn S · S0 where S, Si are sequents and S0 contains exactly one formula. Ex The following rules are focussed. Γ ⇒ A, ∆ Γ ⇒ B, ∆ Γ ⇒ A ∧ B, ∆ Γ, B → C ⇒ A → B Γ, C ⇒ D Γ, (A → B) → C ⇒ D Dfn An axiom is focussed if it is of the form Γ, p ⇒ p, ∆ Γ, ⊥ ⇒ ∆ Γ ⇒ ⊤, ∆ . . .

6 / 12

slide-27
SLIDE 27

Propositional logic Dfn A calculus is terminating if there exists a well-founded order on sequents such that in every rule the premisses come befor the conclusion, and . . .

7 / 12

slide-28
SLIDE 28

Propositional logic Dfn A calculus is terminating if there exists a well-founded order on sequents such that in every rule the premisses come befor the conclusion, and . . . Thm Every terminating calculus that consists of focussed axioms and rules has uniform interpolation.

7 / 12

slide-29
SLIDE 29

Propositional logic Dfn A calculus is terminating if there exists a well-founded order on sequents such that in every rule the premisses come befor the conclusion, and . . . Thm Every terminating calculus that consists of focussed axioms and rules has uniform interpolation. Cor Classical propositional logic has uniform interpolation.

7 / 12

slide-30
SLIDE 30

Propositional logic Dfn A calculus is terminating if there exists a well-founded order on sequents such that in every rule the premisses come befor the conclusion, and . . . Thm Every terminating calculus that consists of focussed axioms and rules has uniform interpolation. Cor Classical propositional logic has uniform interpolation. Cor Intuitionistic propositional logic has uniform interpolation.

7 / 12

slide-31
SLIDE 31

Propositional logic Dfn A calculus is terminating if there exists a well-founded order on sequents such that in every rule the premisses come befor the conclusion, and . . . Thm Every terminating calculus that consists of focussed axioms and rules has uniform interpolation. Cor Classical propositional logic has uniform interpolation. Cor Intuitionistic propositional logic has uniform interpolation. Cor Except the seven intermediate logics that have interpolation, no intermediate logic has a terminating sequent calculus that consists of focussed rules and axioms.

7 / 12

slide-32
SLIDE 32

Propositional logic Dfn A calculus is terminating if there exists an well-founded order on sequents such that in every rule the premisses come befor the conclusion, and . . .

8 / 12

slide-33
SLIDE 33

Propositional logic Dfn A calculus is terminating if there exists an well-founded order on sequents such that in every rule the premisses come befor the conclusion, and . . . Thm Every logic with a terminating calculus that consists of focussed axioms and rules has uniform interpolation.

8 / 12

slide-34
SLIDE 34

Propositional logic Dfn A calculus is terminating if there exists an well-founded order on sequents such that in every rule the premisses come befor the conclusion, and . . . Thm Every logic with a terminating calculus that consists of focussed axioms and rules has uniform interpolation. Proof idea: Define interpolation for rules. For every instance S1 . . . Sn S0 R

  • f a rule, define the formula ∀

R

pS0 in terms of ∀pSi (i > 0). For example, ∀

R

pS0 ≡ ∀pS1 ∧ . . . ∧ ∀pSn.

8 / 12

slide-35
SLIDE 35

Propositional logic Dfn A calculus is terminating if there exists an well-founded order on sequents such that in every rule the premisses come befor the conclusion, and . . . Thm Every logic with a terminating calculus that consists of focussed axioms and rules has uniform interpolation. Proof idea: Define interpolation for rules. For every instance S1 . . . Sn S0 R

  • f a rule, define the formula ∀

R

pS0 in terms of ∀pSi (i > 0). For example, ∀

R

pS0 ≡ ∀pS1 ∧ . . . ∧ ∀pSn. Then inductively define ∀pS ≡

  • {∀

R

pS | R is an instance of a rule with conclusion S}

8 / 12

slide-36
SLIDE 36

Propositional logic Dfn A calculus is terminating if there exists an well-founded order on sequents such that in every rule the premisses come befor the conclusion, and . . . Thm Every logic with a terminating calculus that consists of focussed axioms and rules has uniform interpolation. Proof idea: Define interpolation for rules. For every instance S1 . . . Sn S0 R

  • f a rule, define the formula ∀

R

pS0 in terms of ∀pSi (i > 0). For example, ∀

R

pS0 ≡ ∀pS1 ∧ . . . ∧ ∀pSn. Then inductively define ∀pS ≡

  • {∀

R

pS | R is an instance of a rule with conclusion S} For free sequents S define ∀pS separately.

8 / 12

slide-37
SLIDE 37

Propositional logic Dfn A calculus is terminating if there exists an well-founded order on sequents such that in every rule the premisses come befor the conclusion, and . . . Thm Every logic with a terminating calculus that consists of focussed axioms and rules has uniform interpolation. Proof idea: Define interpolation for rules. For every instance S1 . . . Sn S0 R

  • f a rule, define the formula ∀

R

pS0 in terms of ∀pSi (i > 0). For example, ∀

R

pS0 ≡ ∀pS1 ∧ . . . ∧ ∀pSn. Then inductively define ∀pS ≡

  • {∀

R

pS | R is an instance of a rule with conclusion S} For free sequents S define ∀pS separately. Prove with induction on the order that for all sequents S a uniform interpolant ∀pS exists. ⊣

8 / 12

slide-38
SLIDE 38

Propositional logic Dfn (Γ ⇒ ∆)a = Γ and (Γ ⇒ ∆)s = ∆ and ∀pϕ = ∀p( ⇒ ϕ).

9 / 12

slide-39
SLIDE 39

Propositional logic Dfn (Γ ⇒ ∆)a = Γ and (Γ ⇒ ∆)s = ∆ and ∀pϕ = ∀p( ⇒ ϕ). A logic has uniform interpolation if it satisfies the interpolant properties: (∀l) for all p: ⊢ Sa, ∀pS ⇒ Ss; (∀r) for all p: ⊢ Sl · ( ⇒ ∀pSr) if Sl · Sr is derivable and Sl does not contain p.

9 / 12

slide-40
SLIDE 40

Propositional logic Dfn (Γ ⇒ ∆)a = Γ and (Γ ⇒ ∆)s = ∆ and ∀pϕ = ∀p( ⇒ ϕ). A logic has uniform interpolation if it satisfies the interpolant properties: (∀l) for all p: ⊢ Sa, ∀pS ⇒ Ss; (∀r) for all p: ⊢ Sl · ( ⇒ ∀pSr) if Sl · Sr is derivable and Sl does not contain p. From (∀l) obtain ⊢ ∀pϕ → ϕ.

9 / 12

slide-41
SLIDE 41

Propositional logic Dfn (Γ ⇒ ∆)a = Γ and (Γ ⇒ ∆)s = ∆ and ∀pϕ = ∀p( ⇒ ϕ). A logic has uniform interpolation if it satisfies the interpolant properties: (∀l) for all p: ⊢ Sa, ∀pS ⇒ Ss; (∀r) for all p: ⊢ Sl · ( ⇒ ∀pSr) if Sl · Sr is derivable and Sl does not contain p. From (∀l) obtain ⊢ ∀pϕ → ϕ. From (∀r) obtain that ⊢ ψ → ϕ implies ⊢ ψ → ∀pϕ, if ψ does not contain p, by taking Sl = (ψ ⇒ ) and Sr = ( ⇒ ϕ). Hence ⊢ ψ → ϕ ⇔ ⊢ ψ → ∀pϕ, if ψ does not contain p.

9 / 12

slide-42
SLIDE 42

Propositional logic Dfn (Γ ⇒ ∆)a = Γ and (Γ ⇒ ∆)s = ∆ and ∀pϕ = ∀p( ⇒ ϕ). A logic has uniform interpolation if it satisfies the interpolant properties: (∀l) for all p: ⊢ Sa, ∀pS ⇒ Ss; (∀r) for all p: ⊢ Sl · ( ⇒ ∀pSr) if Sl · Sr is derivable and Sl does not contain p. From (∀l) obtain ⊢ ∀pϕ → ϕ. From (∀r) obtain that ⊢ ψ → ϕ implies ⊢ ψ → ∀pϕ, if ψ does not contain p, by taking Sl = (ψ ⇒ ) and Sr = ( ⇒ ϕ). Hence ⊢ ψ → ϕ ⇔ ⊢ ψ → ∀pϕ, if ψ does not contain p. A logic satisfies the interpolant properties if it satisfies: (1) {Sj · (∀pSj ⇒ ) | 1 ≤ j ≤ n} ⊢ S0 · (∀

R

pS0 ⇒ ). (2) {Sl

j · ( ⇒ ∀pSr j ) | 1 ≤ j ≤ n} ⊢ Sl 0 · ( ⇒ ∀ R

pSr

0).

(3) If Sr

0 is no conclusion of R there exists . . .

9 / 12

slide-43
SLIDE 43

Propositional logic Dfn (Γ ⇒ ∆)a = Γ and (Γ ⇒ ∆)s = ∆ and ∀pϕ = ∀p( ⇒ ϕ). A logic has uniform interpolation if it satisfies the interpolant properties: (∀l) for all p: ⊢ Sa, ∀pS ⇒ Ss; (∀r) for all p: ⊢ Sl · ( ⇒ ∀pSr) if Sl · Sr is derivable and Sl does not contain p. From (∀l) obtain ⊢ ∀pϕ → ϕ. From (∀r) obtain that ⊢ ψ → ϕ implies ⊢ ψ → ∀pϕ, if ψ does not contain p, by taking Sl = (ψ ⇒ ) and Sr = ( ⇒ ϕ). Hence ⊢ ψ → ϕ ⇔ ⊢ ψ → ∀pϕ, if ψ does not contain p. A logic satisfies the interpolant properties if it satisfies: (1) {Sj · (∀pSj ⇒ ) | 1 ≤ j ≤ n} ⊢ S0 · (∀

R

pS0 ⇒ ). (2) {Sl

j · ( ⇒ ∀pSr j ) | 1 ≤ j ≤ n} ⊢ Sl 0 · ( ⇒ ∀ R

pSr

0).

(3) If Sr

0 is no conclusion of R there exists . . .

A logic satisfies the above properties if it has a terminating calculus that consists of focussed axioms and rules. ⊣

9 / 12

slide-44
SLIDE 44

Modal logic Dfn A focussed modal rule is of the form ✷S1 · S0 S2 · ✷S1 · ✷S0 where S1 and S2 consist of multisets and S0 of multisets and exactly one atom.

10 / 12

slide-45
SLIDE 45

Modal logic Dfn A focussed modal rule is of the form ✷S1 · S0 S2 · ✷S1 · ✷S0 where S1 and S2 consist of multisets and S0 of multisets and exactly one atom. Ex The following are focussed modal rules. Γ ⇒ p Π, ✷Γ ⇒ ✷p, Σ RK ✷Γ, p ⇒ ✷∆ Π, ✷Γ, ✷p ⇒ ✷∆, Σ p ⇒ ∆ Π, ✷p ⇒ ✷∆, Σ ROK

10 / 12

slide-46
SLIDE 46

Modal logic Dfn A focussed modal rule is of the form ✷S1 · S0 S2 · ✷S1 · ✷S0 where S1 and S2 consist of multisets and S0 of multisets and exactly one atom. Ex The following are focussed modal rules. Γ ⇒ p Π, ✷Γ ⇒ ✷p, Σ RK ✷Γ, p ⇒ ✷∆ Π, ✷Γ, ✷p ⇒ ✷∆, Σ p ⇒ ∆ Π, ✷p ⇒ ✷∆, Σ ROK Cor A modal logic with a balanced terminating calculus that consists of focussed axioms and focussed (modal) rules and contains RK or ROK, has uniform interpolation.

10 / 12

slide-47
SLIDE 47

Modal logic Dfn A focussed modal rule is of the form ✷S1 · S0 S2 · ✷S1 · ✷S0 where S1 and S2 consist of multisets and S0 of multisets and exactly one atom. Ex The following are focussed modal rules. Γ ⇒ p Π, ✷Γ ⇒ ✷p, Σ RK ✷Γ, p ⇒ ✷∆ Π, ✷Γ, ✷p ⇒ ✷∆, Σ p ⇒ ∆ Π, ✷p ⇒ ✷∆, Σ ROK Cor A modal logic with a balanced terminating calculus that consists of focussed axioms and focussed (modal) rules and contains RK or ROK, has uniform interpolation. Cor Any normal modal logic with a balanced terminating calculus that consists

  • f focussed (modal) axioms and rules, has uniform interpolation. (Ex: K)

10 / 12

slide-48
SLIDE 48

Modal logic Dfn A focussed modal rule is of the form ✷S1 · S0 S2 · ✷S1 · ✷S0 where S1 and S2 consist of multisets and S0 of multisets and exactly one atom. Ex The following are focussed modal rules. Γ ⇒ p Π, ✷Γ ⇒ ✷p, Σ RK ✷Γ, p ⇒ ✷∆ Π, ✷Γ, ✷p ⇒ ✷∆, Σ p ⇒ ∆ Π, ✷p ⇒ ✷∆, Σ ROK Cor A modal logic with a balanced terminating calculus that consists of focussed axioms and focussed (modal) rules and contains RK or ROK, has uniform interpolation. Cor Any normal modal logic with a balanced terminating calculus that consists

  • f focussed (modal) axioms and rules, has uniform interpolation. (Ex: K)

Cor The logics K4 and S4 do not have balanced terminating sequent calculi that consist of focussed (modal) axioms and rules.

10 / 12

slide-49
SLIDE 49

Questions

  • Extend the method to other modal logics, such as GL and KT.
  • Extend the method to hypersequents.
  • Use other proof systems than sequent calculi.

11 / 12

slide-50
SLIDE 50

Finis

12 / 12