bounded arithmetic in free logic
play

Bounded Arithmetic in Free Logic Yoriyuki Yamagata CTFM, 2013/02/20 - PowerPoint PPT Presentation

Bounded Arithmetic in Free Logic Yoriyuki Yamagata CTFM, 2013/02/20 Busss theories 2 Language of Peano Arithmetic + # a # b = 2 | | BASIC axioms PIND , , ( ) 2


  1. Bounded Arithmetic in Free Logic Yoriyuki Yamagata CTFM, 2013/02/20

  2. 𝑗 Buss’s theories 𝑇 2 β€’ Language of Peano Arithmetic + β€œ#” – a # b = 2 𝑏 β‹… | 𝑐 | β€’ BASIC axioms β€’ PIND 𝑦 𝐡 , Ξ“ β†’ Ξ” , 𝐡 ( 𝑦 ) 2 𝐡 0 , Ξ“ β†’ Ξ” , 𝐡 ( 𝑒 ) 𝑐 , i.e. has 𝑗 -alternations of where 𝐡 𝑦 ∈ Ξ£ 𝑗 bounded quantifiers βˆ€π‘¦ ≀ 𝑒 , βˆƒπ‘¦ ≀ 𝑒 .

  3. 𝑗 PH and Buss’s theories 𝑇 2 1 = 𝑇 2 2 = 𝑇 2 3 = … 𝑇 2 Implies π‘ž ) = … 𝑄 = β–‘ ( 𝑂𝑄 ) = β–‘ ( Ξ£ 2 We can approach (non) collapse of PH from (non) collapse of hierarchy of Buss’s theories (PH = Polynomial Hierarchy)

  4. Our approach 𝑗 by GΓΆdel incompleteness theorem β€’ Separate 𝑇 2 β€’ Use analogy of separation of 𝐽Σ 𝑗

  5. Separation of 𝐽Σ 𝑗 … 𝐽Σ 3 ⊒ Con(I Ξ£ 2 ) βŠ† 𝐽Σ 2 ⊒ Con I Ξ£ 2 βŠ† 𝐽Σ 1

  6. 𝑗 Consistency proof inside 𝑇 2 β€’ Bounded Arithmetics generally are not capable to prove consistency. – 𝑇 2 does not prove consistency of Q (Paris, Wilkie) – 𝑇 2 does not prove bounded consistency of 1 (PudlΓ‘k) 𝑇 2 𝑗 does not prove consistency the 𝐢 𝑗 𝑐 fragement – 𝑇 2 βˆ’1 (Buss and IgnjatoviΔ‡ ) of 𝑇 2

  7. Buss and IgnjatoviΔ‡ (1995) … 3 ⊒ 𝐢 3 b βˆ’ Con( 𝑇 2 βˆ’1 ) 𝑇 2 βŠ† 2 ⊒ 𝐢 2 b βˆ’ Con( 𝑇 2 βˆ’1 ) 𝑇 2 βŠ† 1 ⊒ 𝐢 1 b βˆ’ Con( 𝑇 2 βˆ’1 ) 𝑇 2

  8. Where… 𝑐 βˆ’ 𝐷𝐷𝐷 π‘ˆ β€’ 𝐢 𝑗 𝑐 βˆ’ proofs – consistency of 𝐢 𝑗 𝑐 βˆ’ proofs : the proofs by 𝐢 𝑗 𝑐 -formule – 𝐢 𝑗 𝑐 : Ξ£ 0 𝑐 ) … Formulas generated from Ξ£ 𝑗 𝑐 by 𝑐 ( Ξ£ 𝑗 – 𝐢 𝑗 Boolean connectives and sharply bounded quantifiers. βˆ’1 β€’ 𝑇 2 𝑗 – Induction free fragment of 𝑇 2

  9. If… π‘˜ ⊒ 𝐢 i b βˆ’ Con 𝑇 2 βˆ’1 , j > i 𝑇 2 Then, Buss’s hierarchy does not collapse.

  10. βˆ’1 inside 𝑇 2 𝑗 Consistency proof of 𝑇 2 Problem β€’ No truth definition, because β€’ No valuation of terms, because β€’ The values of terms increase exponentially β€’ E.g. 2#2#2#2#2#...#2 𝑗 world, terms do not have values a priori . In 𝑇 2 β€’ Thus, we must prove the existence of values in proofs. β€’ We introduce the predicate 𝐹 which signifies existence of values.

  11. Our result(2012) … 5 ⊒ 3 βˆ’ Con( 𝑇 2 βˆ’1 𝐹 ) 𝑇 2 βŠ† 4 ⊒ 2 βˆ’ Con( 𝑇 2 βˆ’1 𝐹 ) 𝑇 2 βŠ† 3 ⊒ 1 βˆ’ Con( 𝑇 2 βˆ’1 𝐹 ) 𝑇 2

  12. Where… β€’ 𝑗 βˆ’ 𝐷𝐷𝐷 π‘ˆ – consistency of 𝑗 -normal proofs – 𝑗 -normal proofs : the proofs by 𝑗 -normal formulas – 𝑗 -normal formulas: Formulas in the form: βˆƒπ‘¦ 1 ≀ 𝑒 1 βˆ€π‘¦ 2 ≀ 𝑒 2 … 𝑅𝑦 𝑗 ≀ 𝑒 𝑗 𝑅𝑦 𝑗+1 ≀ 𝑒 𝑗+1 . 𝐡 (… ) Where 𝐡 is quantifier free

  13. Where… βˆ’1 𝐹 β€’ 𝑇 2 𝑗 𝐹 – Induction free fragment of 𝑇 2 – have predicate 𝐹 which signifies existence of values β€’ Such logic is called Free logic

  14. 𝑗 𝐹 (Language) 𝑇 2 Predicates β€’ =, ≀ , 𝐹 Function symbols β€’ Finite number of polynomial functions Formulas β€’ Atomic formula, negated atomic formula β€’ 𝐡 ∨ 𝐢 , 𝐡 ∧ 𝐢 β€’ Bounded quantifiers

  15. 𝑗 𝐹 (Axioms) 𝑇 2 β€’ 𝐹 -axioms β€’ Equality axioms β€’ Data axioms β€’ Defining axioms β€’ Auxiliary axioms

  16. Idea behind axioms… β†’ 𝑏 = 𝑏 Because there is no guarantee of 𝐹𝑏 Thus, we add 𝐹𝑏 in the antecedent 𝐹𝑏 β†’ 𝑏 = 𝑏

  17. E-axioms β€’ 𝐹𝐹 𝑏 1 , … , 𝑏 π‘œ β†’ 𝐹𝑏 π‘˜ β€’ 𝑏 1 = 𝑏 2 β†’ 𝐹𝑏 π‘˜ β€’ 𝑏 1 β‰  𝑏 2 β†’ 𝐹𝑏 π‘˜ β€’ 𝑏 1 ≀ 𝑏 2 β†’ 𝐹𝑏 π‘˜ β€’ Β¬ 𝑏 1 ≀ 𝑏 2 β†’ 𝐹𝑏 π‘˜

  18. Equality axioms β€’ 𝐹𝑏 β†’ 𝑏 = 𝑏 β€’ 𝐹𝐹 𝑏 βƒ— , 𝑏 βƒ— = 𝑐 β†’ 𝐹 𝑏 βƒ— = 𝐹 𝑐

  19. Data axioms β€’ β†’ 𝐹𝐹 β€’ 𝐹𝑏 β†’ 𝐹𝑑 0 𝑏 β€’ 𝐹𝑏 β†’ 𝐹𝑑 1 𝑏

  20. Defining axioms 𝐹 𝑣 𝑏 1 , 𝑏 2 , … , 𝑏 π‘œ = 𝑒 ( 𝑏 1 , … , 𝑏 π‘œ ) 𝑣 𝑏 = 0, 𝑏 , 𝑑 0 𝑏 , 𝑑 1 𝑏 𝐹𝑏 1 , … , 𝐹𝑏 π‘œ , 𝐹𝑒 𝑏 1 , … , 𝑏 π‘œ β†’ 𝐹 𝑣 𝑏 1 , 𝑏 2 , … , 𝑏 π‘œ = 𝑒 ( 𝑏 1 , … , 𝑏 π‘œ )

  21. Auxiliary axioms 𝑏 = 𝑐 βŠƒ 𝑏 # 𝑑 = 𝑐 # 𝑑 𝐹𝑏 # 𝑑 , 𝐹𝑐 # 𝑑 , 𝑏 = | 𝑐 | β†’ 𝑏 # 𝑑 = 𝑐 # 𝑑

  22. PIND-rule 𝑐 -formulas where 𝐡 is an Ξ£ 𝑗

  23. 𝑗 𝐹 Bootstrapping 𝑇 2 𝑗 𝐹 ⊒ Tot( 𝐹 ) for any 𝐹 , 𝑗 β‰₯ 0 I. 𝑇 2 𝑗 𝐹 ⊒ BASIC βˆ— , equality axioms βˆ— II. 𝑇 2 𝑗 𝐹 ⊒ predicate logic βˆ— III. 𝑇 2 𝑐 βˆ’PIND βˆ— 𝑗 𝐹 ⊒ Ξ£ 𝑗 IV. 𝑇 2

  24. Theorem (Consistency) 𝑗+2 ⊒ i βˆ’ Con( 𝑇 2 βˆ’1 𝐹 ) 𝑇 2

  25. Valuation trees ρ -valuation tree bounded by 19 ρ(a)=2, ρ(b)=3 a=2 a#a=16 b=3 a#a+b=19 𝑀 𝑏 # 𝑏 + 𝑐 , 𝜍 ↓ 19 19 𝑐 𝑀 𝑒 , 𝜍 ↓ 𝑣 𝑑 is Ξ£ 1

  26. Bounded truth definition (1) β€’ π‘ˆ 𝑣 , 𝑒 1 = 𝑒 2 , 𝜍 ⇔ def βˆƒπ‘‘ ≀ 𝑣 , 𝑀 𝑒 1 , 𝜍 ↓ 𝑣 𝑑 ∧ 𝑀 𝑒 1 , 𝜍 ↓ 𝑣 𝑑 β€’ π‘ˆ 𝑣 , 𝜚 1 ∧ 𝜚 2 , 𝜍 ⇔ def π‘ˆ 𝑣 , 𝜚 1 , 𝜍 ∧ π‘ˆ 𝑣 , 𝜚 2 , 𝜍 β€’ π‘ˆ 𝑣 , 𝜚 1 ∨ 𝜚 2 , 𝜍 ⇔ def π‘ˆ 𝑣 , 𝜚 1 , 𝜍 ∨ π‘ˆ 𝑣 , 𝜚 2 , 𝜍

  27. Bounded truth definition (2) β€’ π‘ˆ 𝑣 , βˆƒπ‘¦ ≀ 𝑒 , 𝜚 ( 𝑦 ) , 𝜍 ⇔ def βˆƒπ‘‘ ≀ 𝑣 , 𝑀 𝑒 , 𝜍 ↓ 𝑣 𝑑 ∧ βˆƒπ‘’ ≀ 𝑑 , π‘ˆ 𝑣 , 𝜚 𝑦 , 𝜍 𝑦 ↦ 𝑒 β€’ π‘ˆ 𝑣 , βˆ€π‘¦ ≀ 𝑒 , 𝜚 ( 𝑦 ) , 𝜍 ⇔ def βˆƒπ‘‘ ≀ 𝑣 , 𝑀 𝑒 , 𝜍 ↓ 𝑣 𝑑 ∧ βˆ€π‘’ ≀ 𝑑 , π‘ˆ ( 𝑣 , 𝜚 𝑦 , 𝜍 [ 𝑦 ↦ 𝑒 ]) 𝑐 , π‘ˆ 𝑣 , 𝜚 is Ξ£ 𝑗+1 𝑐 Remark: If 𝜚 is Ξ£ 𝑗

  28. induction hypothesis 𝑣 : enough large integer 𝑠 : node of a proof of 0=1 Ξ“ 𝑠 β†’ Ξ” 𝑠 : the sequent of node 𝑠 𝜍 : assignment 𝜍 𝑏 ≀ 𝑣 βˆ€π‘£ β€² ≀ 𝑣 βŠ– 𝑠 , { βˆ€π΅ ∈ Ξ“ 𝑠 π‘ˆ 𝑣 β€² , 𝐡 , 𝜍 βŠƒ [ βˆƒπΆ ∈ Ξ” r , π‘ˆ ( 𝑣 β€² βŠ• 𝑠 , 𝐢 , 𝜍 ) ]}

  29. Conjecture βˆ’1 𝐹 is weak enough β€’ 𝑇 2 𝑗+2 can prove 𝑗 -consistency of 𝑇 2 βˆ’1 𝐹 – 𝑇 2 βˆ’1 𝐹 is strong enough β€’ While 𝑇 2 𝑗 𝐹 can interpret 𝑇 2 𝑗 – 𝑇 2 β€’ Conjecture βˆ’1 𝐹 is a good candidate to separate 𝑇 2 𝑗 and 𝑇 2 𝑗+2 . 𝑇 2

  30. Future works β€’ Long-term goal 𝑗 ⊒ π‘—βˆ’Con(𝑇 2 βˆ’1 𝐹 )? 𝑇 2 β€’ Short-term goal 𝑗 𝐹 – Simplify 𝑇 2

  31. Publications β€’ Bounded Arithmetic in Free Logic Logical Methods in Computer Science Volume 8, Issue 3, Aug. 10, 2012

Recommend


More recommend