Reflecting on truth in a partial setting Martin Fischer MCMP LMU - - PowerPoint PPT Presentation

reflecting on truth in a partial setting
SMART_READER_LITE
LIVE PREVIEW

Reflecting on truth in a partial setting Martin Fischer MCMP LMU - - PowerPoint PPT Presentation

Reflecting on truth in a partial setting Martin Fischer MCMP LMU Bristol-M unchen Conference on Truth and Rationality 10. 6. 2016 Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M unchen 1 / 36 Overview The


slide-1
SLIDE 1

Reflecting on truth in a partial setting

Martin Fischer

MCMP LMU

Bristol-M¨ unchen Conference on Truth and Rationality

  • 10. 6. 2016

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 1 / 36

slide-2
SLIDE 2

Overview

The study of reflection principles are important in the arithmetical setting. Also for theories of truth the investigation of reflection principles is important and fruitful. What about reflection principles in a partial setting? What about the connection between reflection and PKF?

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 2 / 36

slide-3
SLIDE 3

Content

1

Background Axiomatizing Kripke N-Categoricity Infinitary proof systems

2

Reflection From the ω-rule to reflection From Tarski biconditionals to KF

3

Reflecting on truth in a partial setting Partial logic Recovering PKF Induction

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 3 / 36

slide-4
SLIDE 4

Background Axiomatizing Kripke

Kripke models

Kripke: Fixed-point construction for different evaluation schemes e. monotone operators Γe. Fixed-points Γe(S) = S for S ⊆ N. Focus: strong Kleene, e = sk. The minimal fixed-point for strong Kleene Isk.

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 4 / 36

slide-5
SLIDE 5

Background Axiomatizing Kripke

Axiomatizing Kripke

Axiomatizations: KF (Feferman) The problem of external and internal logic. IKF (Reinhardt) ({A ∈ LT | KF ⊢ T(A)}) The problem of natural axiomatization. PKF (Halbach/Horsten) In what sense are these axiomatizations and which one is preferable?

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 5 / 36

slide-6
SLIDE 6

Background N-Categoricity

N-Categoricity

Suggestion: N-categoricity. Fix the interpretation of the arithmetical part with the standard model N. Σ is N-categorical for a set of models M iff (N, S) | = Σ ⇔ S ∈ M For the minimal fixed-point: (N, S) | = Σ ⇔ S = Isk For arbitrary fixed-points: (N, S) | = Σ ⇔ S = Γsk(S)

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 6 / 36

slide-7
SLIDE 7

Background N-Categoricity

N-Categoricity

The minimal fixed-point is Π1

1-complete (Kripke, Burgess).

There is no N-categorical axiomatization of the minimal fixed-point. KF is an N-categorical axiomatization of arbitrary fixed-points. (Feferman) TFB is an N-categorical axiomatization of arbitrary fixed-points. (Leigh) IKF is not N-categorical axiomatization of arbitrary fixed-points. Conclusion: KF is at best an axiomatization of arbitrary fixed-points and N-categoricity cannot be the only criterion.

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 7 / 36

slide-8
SLIDE 8

Background N-Categoricity

N-Categoricity and partiality

The set of derivable sequents of PKF is an N-categorical axiomatization of arbitrary fixed-points. The set of theorems of PKF, i.e. sequents of the form ⇒ A, is not N-categorical axiomatization of arbitrary fixed-points. The set of truth sequents T(A) ⇒ A, A ⇒ T(A) is an N-categorical axiomatization of arbitrary fixed-points.

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 8 / 36

slide-9
SLIDE 9

Background Infinitary proof systems

Infinitary proof systems

Infinitary proof systems allow for characterizations of the minimal fixed-points. Cantini has an infinitary proof system (sequent system with ω-rule) characterizing the minimal fixed-point of supervaluation. Welch gametheoretic characterization. Meadows infinitary tableaux.

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 9 / 36

slide-10
SLIDE 10

Background Infinitary proof systems

Infinitary proof system for strong Kleene

Example SK∞ a Tait system: Initial sequents ⇒ A (for true atomic arithmetical sentences) ⇒ A ⇒ Γ, T(A) ⇒ ¬A ⇒ Γ, ¬T(A) ... A(n) ... ω-rule (for all n ∈ N) ∀xA(x) Then SK∞ ⊢ A ⇔ #A ∈ Isk

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 10 / 36

slide-11
SLIDE 11

Background Infinitary proof systems

Embeddings into infinitary proof systems

Similar to the Gentzen-Sch¨ utte method we can look at embeddings into the infinitary proof systems. KF cannot be directly embedded. An embedding of the theorems of PKF into SK∞ is possible

◮ if PKF ⊢⇒ A, then #A ∈ Γωω (Cantini, Halbach/Horsten). ◮ for the language of truth we only have transfinite induction up to ωω in

PKF.

IKF is contained in Isk

◮ if IKF ⊢ A, then #A ∈ Γǫ0 (Cantini). ◮ for the language of truth we have transfinite induction up to ǫ0 in KF. Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 11 / 36

slide-12
SLIDE 12

Reflection From the ω-rule to reflection

Uniform reflection as a finitary ω-rule

∀xPrΣ(A˙ x) (RFNR

Σ)

∀xA(x) (RFNΣ) ∀x(PrΣ(A˙ x) → A(x)). Hilbert 1931. Shoenfield constructivized version of the ω-rule. Feferman 1962 showed the equivalence.

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 12 / 36

slide-13
SLIDE 13

Reflection From the ω-rule to reflection

The strength of uniform reflection

For an axiomatizable theory Σ we use R(Σ) := EAT + RFNΣ. TB0 is EAT+ Tarski biconditionals for sentences of LA. UTB0 is EAT+ uniform Tarski biconditionals for formulas of LA. TFB0 is EAT+ truth and falsity biconditionals for sentences of LP, i.e. the language of we get by adding F as the dual for T and allow

  • nly positive occurrences of T and F.

T(A) ↔ A & F(A) ↔ A UTFB0 is EAT+ uniform truth and falsity biconditionals for formulas

  • f LP.

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 13 / 36

slide-14
SLIDE 14

Reflection From Tarski biconditionals to KF

Truth and Reflection

Reflecting on Tarski biconditionals gives uniform Tarski biconditionals.

Lemma (Horsten, Leigh)

UTB0 ⊆ R(TB0). Reflecting on typefree truth and falsity biconditionals gives uniform typefree truth and falsity biconditionals.

Lemma (Horsten, Leigh)

UTFB0 ⊆ R(TFB0).

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 14 / 36

slide-15
SLIDE 15

Reflection From Tarski biconditionals to KF

Truth and Reflection

Reflecting on uniform Tarski biconditionals gives the compositional axioms.

Lemma (Halbach)

CT0 ⊆ R(UTB0). Reflecting on uniform truth and falsity biconditionals gives the compositional axioms of KF.

Lemma (Horsten, Leigh)

KF ⊆ R(UTFB0).

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 15 / 36

slide-16
SLIDE 16

Reflecting on truth in a partial setting Partial logic

Partial logic

The logic is four valued. Gaps and gluts. Logical consequence for sequents:

◮ Truth preservation ◮ Falsity antipreservation Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 16 / 36

slide-17
SLIDE 17

Reflecting on truth in a partial setting Partial logic

Basic

For negation we have contraposition Γ ⇒ ∆ ¬∆ ⇒ ¬Γ but not A, Γ ⇒ ∆ Γ ⇒ ∆, ¬A Γ ⇒ ∆, A ¬A, Γ ⇒ ∆ We assume as background an arithmetical theory BASIC formulated in LT: EAT formulated in a sequent version of partial logic along the lines of Halbach 2014.

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 17 / 36

slide-18
SLIDE 18

Reflecting on truth in a partial setting Partial logic

Minimal truth TS0

TS0 is obtained by extending BASIC with the initial sequents T1 T(A) ⇒ A T2 A ⇒ T(A) Simplicity. No need for restriction of the language.

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 18 / 36

slide-19
SLIDE 19

Reflecting on truth in a partial setting Partial logic

Reflection as a rule

Assume some coding of finite sets of formulas [Γ], then [Γ˙ x] denotes the result of substituting in Γ the x-th numeral for x. [Γ˙ x] ⇒ . [∆˙ x] denotes the sequent Γ(x) ⇒ ∆(x) with the possible free variable x and the dots indicate as usual the use of the sub and num function. Let Σ be an axiomatizable theory, then R(Σ) = EAT + RFNR

Σ.

PrΣ([Γ˙ x] ⇒ . [∆˙ x]) (RFNR

Σ)

Γ(x) ⇒ ∆(x)

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 19 / 36

slide-20
SLIDE 20

Reflecting on truth in a partial setting Recovering PKF

From TS0 to UTS0

R(TS0) ⊢

(i) A(x) ⇒ T(A˙ x); (ii) T(A˙ x) ⇒ A(x). Argument: For all formulas A(x) and for all n ∈ N: TS0 ⊢ A(n) ⇒ T(A(n)). As this is uniform we get in the formalization EAT ⊢⇒ PrTS0([A˙ x] ⇒ . [T(A)˙ x]). With reflection we get R(TS0) ⊢ A(x) ⇒ T(A˙ x).

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 20 / 36

slide-21
SLIDE 21

Reflecting on truth in a partial setting Recovering PKF

Regaining compositional sequents I

R(TS0) ⊢

(i) sent(x), sent(y), T(x ∧ . y) ⇒ T(x) ∧ T(y); (ii) sent(x), sent(y), T(x) ∧ T(y) ⇒ T(x ∧ . y); (iii) sent(x), sent(y), T(x ∨ . y) ⇒ T(x) ∨ T(y); (iv) sent(x), sent(y), T(x) ∨ T(y) ⇒ T(x ∨ . y); (v) sent(x), ¬T(x) ⇒ T( ¬ . x); (vi) sent(x), T( ¬ . x) ⇒ ¬T(x).

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 21 / 36

slide-22
SLIDE 22

Reflecting on truth in a partial setting Recovering PKF

Regaining compositional sequents II

R(UTS0) ⊢

(i) sent( ∀ . xy), ∀xT(y ˙ x) ⇒ T( ∀ . xy); (ii) sent( ∀ . xy), T( ∀ . xy) ⇒ ∀xT(y ˙ x); (iii) sent( ∃ . xy), ∃xT(y ˙ x) ⇒ T( ∃ . xy); (iv) sent( ∃ . xy), T( ∃ . xy) ⇒ ∃xT(y ˙ x).

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 22 / 36

slide-23
SLIDE 23

Reflecting on truth in a partial setting Recovering PKF

Regaining compositional sequents III

R(UTS0) ⊢

(i) ct(x), T(val(x)) ⇒ T( T . x); (ii) ct(x), T( T . x) ⇒ T(val(x)); (iii) ct(x), ct(y), val(x) = val(y) ⇒ T(x = . y); (iv) ct(x), ct(y), T(x = . y) ⇒ val(x) = val(y).

Observation

PKF0 ⊆ R(UTS0) ⊂ R(R(TS0))

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 23 / 36

slide-24
SLIDE 24

Reflecting on truth in a partial setting Induction

Induction in classical arithmetic

Theorem (Kreisel and L´ evy)

R(EA) = PA. Argument for ⊇: For a formula A with one free variable let B(x) be A(0) ∧ ∀x(A(x) → A(x + 1)) → A(x). Then we can argue in EA by external induction that for all k, EA ⊢ B(k). Since the size of the proofs can be bound by an elementary function we can formalize the induction in

  • EA. So we get EA ⊢ PrEA(B ˙

x) and with reflection B(x). Similarly we get R(EAT) = PAT.

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 24 / 36

slide-25
SLIDE 25

Reflecting on truth in a partial setting Induction

Induction for LT (partial)

Instead of using the (schema) of induction, the following rule is adopted: A(x), Γ ⇒ ∆, A(x + 1) (Ind) A(0), Γ ⇒ ∆, A(t) In R(UTS0) we get induction for all formulas of LT and so

Observation

PKF ⊆ R(UTS0)) ⊂ R(R(TS0)).

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 25 / 36

slide-26
SLIDE 26

Reflecting on truth in a partial setting Induction

Transfinite induction

For a fixed ordinal representation, for example with the Cantor normal form for ordinals < ǫ0 we define:

Definition

Let A be a formula with one free variable Prog(A) := ∀α < βA(α) → A(β). TI(A, β) := Prog(A) → ∀α < βA(α). TIL(< α) := {TI(A, β) | A ∈ L & β < α}.

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 26 / 36

slide-27
SLIDE 27

Reflecting on truth in a partial setting Induction

Transfinite induction for a language with truth

Lemma

Reflecting on EAT gives TILT (< ǫ0). Argument: Similar to PA proves transfinite induction up to ǫ0. For a formula A(x) define A′(x) to be ∀β(∀α < βA(α) → ∀α < β + ωxA(α)) Then we show Prog(A) → Prog(A′). With this TI(A, α) ⇒ TI(A, ωα), and finally TILT (< ǫ0).

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 27 / 36

slide-28
SLIDE 28

Reflecting on truth in a partial setting Induction

TILT in a partial setting

Prog(A) := ∀α < βA(α) ⇒ A(β) Prog(A) TIR(A, β) ⇒ ∀α < βA(α) TIRLT (< α) is the closure under the rules TIR(A, β) for all A ∈ LT and for all β < α.

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 28 / 36

slide-29
SLIDE 29

Reflecting on truth in a partial setting Induction

TIRLT(< ǫ0) in R(UTS)?

Basic proof strategy: Show Prog(A) Prog(A′) then closure under TIR(A, β) implies closure under TIR(A, ωβ) for all A ∈ LT.

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 29 / 36

slide-30
SLIDE 30

Reflecting on truth in a partial setting Induction

Problems for the direct argument

We run into problems if we try to show that Prog(A) Prog(A′) Remember that A′(x) is ∀β(∀α < βA(α) → ∀α < β + ωxA(α)). In our partial setting we do not have in general ⇒ A ⇒ A → B ⇒ B

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 30 / 36

slide-31
SLIDE 31

Reflecting on truth in a partial setting Induction

Idea

Idea (Carlo): circumvent the MP argument step. In UTS we can prove (by external induction) for all n that Prog(A) ∀α < βA(α) ⇒ ∀α < β + ωnA(α) Problem: How to use this fact?

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 31 / 36

slide-32
SLIDE 32

Reflecting on truth in a partial setting Induction

Reflection on rules

Solution: Strengthening of reflection. Assume that Σ allows for the following derivation Γ ⇒ ∆ Θ ⇒ Λ Then a reflection on Σ should also include this fact PrΣ([Γ˙ x] ⇒ . [∆˙ x], [Θ˙ x] ⇒ . [Λ˙ x]) Γ(x) ⇒ ∆(x) (R∗) Θ(x) ⇒ Λ(x)

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 32 / 36

slide-33
SLIDE 33

Reflecting on truth in a partial setting Induction

TIRLT(< ǫ0) in R∗(UTS)

Now we can formalize the external induction to get PrUTS([Prog(A)], [∀α < βA(α)] ⇒ . [∀α < β + ωxA(α)˙ x]) and with reflection we have Prog(A) ∀α < βA(α) ⇒ ∀α < β + ωxA(α) Setting β = 0 we can then argue for TIRLT (< ǫ0).

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 33 / 36

slide-34
SLIDE 34

Reflecting on truth in a partial setting Induction

Open questions

In classic theories we have a close connection between reflection and induction. Is it as close in partial logic? Is reflection able to close the (proof theoretic) gap between PKF and IKF?

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 34 / 36

slide-35
SLIDE 35

Reflecting on truth in a partial setting Induction

Concluding remarks

Theories of truth built on reflection principles are very well motivated. Reflection on simple truth sequents allows us to gain the compositional axioms of PKF. Reflection and induction are closely connected also in the partial setting. Reflection gives full induction. Reflection gives TIRLT (< ǫ0).

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 35 / 36

slide-36
SLIDE 36

Reflecting on truth in a partial setting Induction

Thank you!

Martin Fischer (MCMP LMU) Reflecting on truth in a partial setting Bristol-M¨ unchen 36 / 36