Effective computability and constructive provability for existence - - PowerPoint PPT Presentation

effective computability and constructive provability for
SMART_READER_LITE
LIVE PREVIEW

Effective computability and constructive provability for existence - - PowerPoint PPT Presentation

Effective computability and constructive provability for existence sentences Makoto Fujiwara Waseda Institute for Advanced Study (WIAS), Waseda University Workshop on Mathematical Logic and its Application, Kyoto University, 16 September 2016


slide-1
SLIDE 1

Effective computability and constructive provability for existence sentences

Makoto Fujiwara

Waseda Institute for Advanced Study (WIAS), Waseda University

Workshop on Mathematical Logic and its Application, Kyoto University, 16 September 2016

This work is supported by JSPS KAKENHI Grant Number JP16H07289 and also by JSPS Core-to-Core Program (A. Advanced Research Networks).

1 / 16

slide-2
SLIDE 2

Targets: Existence Statements

Many theorems in ordinary mathematics (Analysis, Algebra, Combinatorics etc.) can be formalized as Π1

2 sentences having

a form ∀f (ϕ(f ) → ∃gψ(f , g)) , where f and g are (possibly tuples of) functions on natural numbers.

2 / 16

slide-3
SLIDE 3

In computable (recursive) mathematics, effective contents

  • f classical theorems have been investigated.

In particular, there are many “effectivized” results of classical theorems in combinatorics, e.g.,

Brooks’s theorem (Schmerl 1982, Carstens/P¨ appinghaus 1983, Tverberg 1984), Marriage theorem (Kierstead 1983), Dilworth’s theorem (Kierstead 1981).

3 / 16

slide-4
SLIDE 4

In computable (recursive) mathematics, effective contents

  • f classical theorems have been investigated.

In particular, there are many “effectivized” results of classical theorems in combinatorics, e.g.,

Brooks’s theorem (Schmerl 1982, Carstens/P¨ appinghaus 1983, Tverberg 1984), Marriage theorem (Kierstead 1983), Dilworth’s theorem (Kierstead 1981).

Two Kinds of Effectivization Non-uniform computability: For any computable f , there exists a computable g. Uniform computability: There exists a uniform algorithm to obtain a witness g for each (computable) f .

3 / 16

slide-5
SLIDE 5

Correspondence in Reverse Mathematics

Non-uniform computability (For any computable f , there exists a computable g.) Uniform computability (There exists a uniform algorithm to ob- tain a witness g for each f .) Constructive provability

4 / 16

slide-6
SLIDE 6

Correspondence in Reverse Mathematics

Non-uniform computability ≈ RCA0 (For any computable f , there exists a computable g.) Uniform computability (There exists a uniform algorithm to ob- tain a witness g for each f .) Constructive provability ≈ EL0

4 / 16

slide-7
SLIDE 7

Correspondence in Reverse Mathematics

Non-uniform computability ≈ RCA0 (For any computable f , there exists a computable g.) Uniform computability ≈ ?? (There exists a uniform algorithm to ob- tain a witness g for each f .) Constructive provability ≈ EL0 Toward an axiomatization of uniform computability Investigate how uniform computability for existence statements can be captured by (semi-)intuitionistic provability in (many-sorted) arithmetic!

4 / 16

slide-8
SLIDE 8

Formalization of Uniform Computability (F. 2015)

Hilbert-type system E-HAω (resp. E-PAω) is the finite type extension of HA (resp. PA), of which T is the terms.

5 / 16

slide-9
SLIDE 9

Formalization of Uniform Computability (F. 2015)

Hilbert-type system E-HAω (resp. E-PAω) is the finite type extension of HA (resp. PA), of which T is the terms.

  • E-HA

ω↾ (resp.

E-PA

ω↾) is the restrictions of E-HAω (resp.

E-PAω) to primitive recursion of type 0 and quantifier-free induction, of which T0 is the terms.

5 / 16

slide-10
SLIDE 10

Formalization of Uniform Computability (F. 2015)

Hilbert-type system E-HAω (resp. E-PAω) is the finite type extension of HA (resp. PA), of which T is the terms.

  • E-HA

ω↾ (resp.

E-PA

ω↾) is the restrictions of E-HAω (resp.

E-PAω) to primitive recursion of type 0 and quantifier-free induction, of which T0 is the terms.

Intuitionistic Logic Classical Logic HA PA 1 EL0 EL RCA0 RCA ω

  • E-HA

ω↾ + QF-AC1,0

E-HAω + QF-AC1,0 RCAω RCAω

5 / 16

slide-11
SLIDE 11

Formalization of Uniform Computability (F. 2015)

Hilbert-type system E-HAω (resp. E-PAω) is the finite type extension of HA (resp. PA), of which T is the terms.

  • E-HA

ω↾ (resp.

E-PA

ω↾) is the restrictions of E-HAω (resp.

E-PAω) to primitive recursion of type 0 and quantifier-free induction, of which T0 is the terms.

Intuitionistic Logic Classical Logic HA PA 1 EL0 EL RCA0 RCA ω

  • E-HA

ω↾ + QF-AC1,0

E-HAω + QF-AC1,0 RCAω RCAω

  • Fact. (Kohlenbach 2005)

RCAω

0 is a conservative extension of RCA0.

5 / 16

slide-12
SLIDE 12

Formalization of Uniform Computability (F. 2015)

Hilbert-type system E-HAω (resp. E-PAω) is the finite type extension of HA (resp. PA), of which T is the terms.

  • E-HA

ω↾ (resp.

E-PA

ω↾) is the restrictions of E-HAω (resp.

E-PAω) to primitive recursion of type 0 and quantifier-free induction, of which T0 is the terms.

Intuitionistic Logic Classical Logic HA PA 1 EL0 EL RCA0 RCA ω

  • E-HA

ω↾ + QF-AC1,0

E-HAω + QF-AC1,0 RCAω RCAω

  • Fact. (Kohlenbach 2005)

RCAω

0 is a conservative extension of RCA0.

That is also the case for WKLω

0 (:= RCAω 0 + WKL) and

ACAω

0 (:= RCAω 0 + ACA).

5 / 16

slide-13
SLIDE 13

Uniform Provability in Γ:

1 There exists a term t1 s.t.

Γ ⊢ ∀f (ϕ(f ) → t|f ↓ ∧ψ(f , t|f )) ,

where α(β) := { α(¯ βn) − 1 where n is the least n′ s.t. α(¯ βn′) ̸= 0. ↑ if there is no such n′. α|β := λn. α(⟨n⟩⌢β).

2 There exists a (G¨

  • del prim. rec.) term t1→1 ∈ T s.t.

Γ ⊢ ∀f (ϕ(f ) → ψ(f , tf )) .

3 There exists a (Kleene prim. rec.) term t1→1 ∈ T0 s.t.

Γ ⊢ ∀f (ϕ(f ) → ψ(f , tf )) .

6 / 16

slide-14
SLIDE 14

Proposition 1. (F. 2015) Let ∀f (ϕ(f ) → ∃gψ(f , g)) be a L(EL0)-formula such that ϕ(f ) is purely universal and ψ(f , g) is equivalent to some formula ∀w ρ∃s0ψqf (f , g, w, s) over EL0. There exists a term t1 such that RCA0(+WKL) ⊢ ∀f (ϕ(f ) → t|f ↓ ∧ψ(f , t|f )) if and only if EL0 ⊢ ∀f (ϕ(f ) → ∃gψ(f , g)) .

7 / 16

slide-15
SLIDE 15

Proposition 1. (F. 2015) Let ∀f (ϕ(f ) → ∃gψ(f , g)) be a L(EL0)-formula such that ϕ(f ) is purely universal and ψ(f , g) is equivalent to some formula ∀w ρ∃s0ψqf (f , g, w, s) over EL0. There exists a term t1 such that RCA0(+WKL) ⊢ ∀f (ϕ(f ) → t|f ↓ ∧ψ(f , t|f )) if and only if EL0 ⊢ ∀f (ϕ(f ) → ∃gψ(f , g)) . On the Proof. IF direction is by the function realizability (Dorais 2014). ONLY IF direction is by Kuroda’s negative translation and the (monotone) Dialectica interpretation.

7 / 16

slide-16
SLIDE 16

Application. For example, Kierstead’s effective marriage theorem EMT has the required syntactical form in Proposition 1 and uniformly provable in RCA0, then it follows that EMT is provable in EL0. Remark. It is known that many existence theorems are formalized as a Π1

2 formula of the syntactical form in Proposition 1.

The analogous results for EL, RCA instead of EL0, RCA0 also hold.

8 / 16

slide-17
SLIDE 17

The Next Step

Most of discussion in ordinary mathematics is carried out in the presence of Arithmetical Comprehension Axiom.

9 / 16

slide-18
SLIDE 18

The Next Step

Most of discussion in ordinary mathematics is carried out in the presence of Arithmetical Comprehension Axiom. Π0

1-AC0,0 :

∀α1(∀x0∃y 0∀z0α(x, y, z) = 0 → ∃β1∀x, z α(x, β(x), z) = 0) classically derives ACA.

9 / 16

slide-19
SLIDE 19

The Next Step

Most of discussion in ordinary mathematics is carried out in the presence of Arithmetical Comprehension Axiom. Π0

1-AC0,0 :

∀α1(∀x0∃y 0∀z0α(x, y, z) = 0 → ∃β1∀x, z α(x, β(x), z) = 0) classically derives ACA. Question. How is uniform provability in classical systems with Π0

1-AC0,0

characterized?

9 / 16

slide-20
SLIDE 20

The Next Step

Most of discussion in ordinary mathematics is carried out in the presence of Arithmetical Comprehension Axiom. Π0

1-AC0,0 :

∀α1(∀x0∃y 0∀z0α(x, y, z) = 0 → ∃β1∀x, z α(x, β(x), z) = 0) classically derives ACA. Question. How is uniform provability in classical systems with Π0

1-AC0,0

characterized? Fact. The negative translation of Π0

1-AC0,0 is intuitionistically

derived from Π0

1-AC0,0 and Σ0 2-DNS0:

∀α1(∀x0¬¬∃y 0∀z0α(x, y, z) = 0 → ¬¬∀x∃y∀zα(x, y, z) = 0).

9 / 16

slide-21
SLIDE 21

Uniform ⇒ Intuitionistic

Proposition 2. Let ∀f (ϕ(f ) → ∃gψ(f , g)) be a L(EL0)-formula such that ϕ(f ) ∈ A and ψ(f , g) ∈ B. If there exists a term t1→1 ∈ T0 such that

  • E-PA

ω↾ + Π0 1-AC0,0 ⊢ ∀f (ϕ(f ) → ψ(f , tf )) ,

then EL0 + Π0

1-AC0,0 + Σ0 2-DNS0 ⊢ ∀f (ϕ(f ) → ∃gψ(f , g)) .

10 / 16

slide-22
SLIDE 22

Uniform ⇒ Intuitionistic

Proposition 2. Let ∀f (ϕ(f ) → ∃gψ(f , g)) be a L(EL0)-formula such that ϕ(f ) ∈ A and ψ(f , g) ∈ B. If there exists a term t1→1 ∈ T0 such that

  • E-PA

ω↾ + Π0 1-AC0,0 ⊢ ∀f (ϕ(f ) → ψ(f , tf )) ,

then EL0 + Π0

1-AC0,0 + Σ0 2-DNS0 ⊢ ∀f (ϕ(f ) → ∃gψ(f , g)) .

The classes A, B of formulas are defined simultaneously by P, A1 ∧ A2, A1 ∨ A2, ∀xA1, ∃xA1, B1 → A1 are in A; P, B1 ∧ B2, ∀xB1, A1 → B1 are in B; where P, Ai, Bi range over prime formulas, formulas in A, B respectively.

10 / 16

slide-23
SLIDE 23

Uniform ⇐ Intuitionistic

  • Proposition. (Hirst/Mummer 2011)

Let ∀f (ϕ(f ) → ∃gψ(f , g)) be a L( E-HA

ω↾)-formula such that

ϕ(f ) is ∃-free and ψ(f , g) ∈ Γ1. If

  • E-HA

ω↾ + AC + IPω ef ⊢ ∀f (ϕ(f ) → ∃gψ(f , g)),

then there exists a term t1→1 ∈ T0 such that

  • E-HA

ω↾ ⊢ ∀f (ϕ(f ) → ψ(f , tf )) .

AC: the axiom of choice in all finite types. IPω

ef: independence of premise scheme for ∃-free formulas.

The proof is by the modified realizability interpretation.

11 / 16

slide-24
SLIDE 24

Uniform ⇐ Intuitionistic

Lemma. The modified realizability interpretation of Σ0

2-DNS0 is

intuitionistically derived from Σ0

2-DNS0 and Π0 1-AC0,0.

12 / 16

slide-25
SLIDE 25

Uniform ⇐ Intuitionistic

Lemma. The modified realizability interpretation of Σ0

2-DNS0 is

intuitionistically derived from Σ0

2-DNS0 and Π0 1-AC0,0.

Proposition 3. Let ∀f (ϕ(f ) → ∃gψ(f , g)) be a L( E-HA

ω↾)-formula such that

ϕ(f ) is ∃-free and ψ(f , g) ∈ Γ1. If

  • E-HA

ω↾ + AC + IPω ef + Σ0 2-DNS0 ⊢ ∀f (ϕ(f ) → ∃gψ(f , g)),

then there exists a term t1→1 ∈ T0 such that

  • E-PA

ω↾ + Π0 1-AC0,0 ⊢ ∀f (ϕ(f ) → ψ(f , tf )) .

12 / 16

slide-26
SLIDE 26

Combining Proposition 2 and 3

13 / 16

slide-27
SLIDE 27

Combining Proposition 2 and 3

Lemma.

1 ∃-free = A ∩ B. 2 B ∩ Γ1 = C.

The class C of formulas is defined by P, C1 ∧ C2, ∀xC1, Q → C1 are in C where P, Q, Ci range over prime formulas, ∃-free formulas, formulas in C respectively.

13 / 16

slide-28
SLIDE 28

Combining Proposition 2 and 3

Lemma.

1 ∃-free = A ∩ B. 2 B ∩ Γ1 = C. 3 C = ∃-free.

The class C of formulas is defined by P, C1 ∧ C2, ∀xC1, Q → C1 are in C where P, Q, Ci range over prime formulas, ∃-free formulas, formulas in C respectively.

13 / 16

slide-29
SLIDE 29

Combining Proposition 2 and 3

Lemma.

1 ∃-free = A ∩ B. 2 B ∩ Γ1 = C. 3 C = ∃-free.

The class C of formulas is defined by P, C1 ∧ C2, ∀xC1, Q → C1 are in C where P, Q, Ci range over prime formulas, ∃-free formulas, formulas in C respectively. Proposition. A ∩ ∃-free = ∃-free. B ∩ Γ1 = ∃-free.

13 / 16

slide-30
SLIDE 30

Theorem. Let ∀f (ϕ(f ) → ∃gψ(f , g)) be a L(EL0)-formula such that ϕ(f ) and ψ(f , g) are ∃-free. There exists exists a term t1→1 ∈ T0 such that

  • E-PA

ω↾ + Π0 1-AC0,0 ⊢ ∀f (ϕ(f ) → ψ(f , tf ))

if and only if EL0 + AC0,0 + Σ0

2-DNS0 ⊢ ∀f (ϕ(f ) → ∃gψ(f , g)) .

14 / 16

slide-31
SLIDE 31

Theorem. Let ∀f (ϕ(f ) → ∃gψ(f , g)) be a L(EL0)-formula such that ϕ(f ) and ψ(f , g) are ∃-free. There exists exists a term t1→1 ∈ T0 such that

  • E-PA

ω↾ + Π0 1-AC0,0 ⊢ ∀f (ϕ(f ) → ψ(f , tf ))

if and only if EL0 + AC0,0 + Σ0

2-DNS0 ⊢ ∀f (ϕ(f ) → ∃gψ(f , g)) .

Remark. The analogous result for T, E-PAω, EL instead of T0, E-PA

ω↾,

EL0 also holds.

14 / 16

slide-32
SLIDE 32

Appendix

Proposition. Let ∀f (ϕ(f ) → ∃gψ(f , g)) be a L(EL0)-formula such that ϕ(f ) is purely universal and ψ(f , g) is equivalent to ∀w ρ∃sτψqf (f , g, w, s) over EL0 (ρ, τ ∈ {0, 1}). There exists a term t1→1 ∈ T0 + B0,1 such that

  • E-PA

ω↾ + Π0 1-AC0,0 + BR0,1 ⊢ ∀f (ϕ(f ) → ψ(f , tf ))

if and only if EL0+AC0,0+MP+Σ0

2-DNS0+BR0,1 ⊢ ∀f (ϕ(f ) → ∃gψ(f , g)) .

  • Fact. (Kohlenbach 1999)

t1→1 ∈ T0 + BR0,1 ⇔ t1→1 ∈ T.

15 / 16

slide-33
SLIDE 33

References

1 M. Fujiwara, “Intuitionistic provability versus uniform

provability in RCA”, Lecture Notes in Computer Science, vol. 9136, pp. 186–195, 2015.

2 F. G. Dorais, “Classical consequences of continuous choice

principles from intuitionistic analysis”, Notre Dame Journal of Formal Logic 55, no.1, pp. 25–39, 2014.

3 J. L. Hirst and C. Mummert, “Reverse mathematics and

uniformity in proofs without excluded middle”, Notre Dame Journal of Formal Logic 52, no.2, pp. 149–162, 2011.

Thank you for your attention!

16 / 16