Download slides

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Structure (mathematical logic) wikipedia , lookup

Propositional calculus wikipedia , lookup

Laws of Form wikipedia , lookup

Transcript
Quantifiers in nonclassical logics
Rosalie Iemhoff
Utrecht University
Algebra and Coalgebra meets Proof Theory
Vienna, 7–9 April 2016
1 / 23
The seventh edition of ALCOP
On 21 July 2009 Nick Bezhanishvili and Clemens Kupke wrote to
Alessandra Palmigiano and me:
“Dear Alessandra, Rosalie,
We recently learned that there is funding available for small scale
workshops between the UK and the Netherlands.
We thought this is a good opportunity to get together - ‘strengthen our
already well-established contacts’ to use grant application vocabulary :)”
What’s in a name: First named ACPT, soon thereafter ALCOP.
2 / 23
Quantifiers versus functions
Ex If ∀x∃y ϕ(x, y ) holds, then introduce a fresh function f and add as
axiom
∀xϕ(x, fx).
This is a conservative extension:
`CQC ∀xϕ(x, fx) → ψ ⇔ `CQC ∀x∃y ϕ(x, y ) → ψ.
3 / 23
Skolemization Method
Thm (Skolem 1920) `CQC ∃x∀y ϕ(x, y ) ⇔ `CQC ∃xϕ x, f (x)
`CQC ∃x1 ∀y1 ∃x2 ∀y2 ϕ(x̄, y1 , y2 ) ⇔ `CQC ∃x1 ∃x2 ϕ x̄, f1 (x1 ), f2 (x1 , x2 )
In general:
`CQC ∃x1 ∀y1 . . . ∃xn ∀yn ϕ(x̄, y1 , . . . , yn )
⇔
`CQC ∃x1 . . . ∃xn ϕ x̄, f1 (x1 ), f2 (x1 , x2 ), . . . , fn (x1 , . . . , xn ) .
⇔
S
`CQC ∃x1 ∀y1 . . . ∃xn ∀yn ϕ(x̄, ȳ )
4 / 23
Herbrand’s Theorem
Thm (Herbrand) For every quantifier-free ϕ:
`CQC ∃x̄ϕ(x̄) ⇔ for some terms t11 , . . . , tnk : `CQC
k
_
ϕ(t1i , . . . , tni ).
i=1
Cor For every quantifier-free ϕ:
`CQC ∃x1 ∀y1 . . . ∃xn ∀yn ϕ(x̄, ȳ )
⇔
`CQC ∃x1 . . . ∃xn ϕ x̄, f1 (x1 ), f2 (x1 , x2 ), . . . , fn (x1 , . . . , xn )
⇔
for some terms t11 , . . . , tnk :
`CPC
Wk
i=1
ϕ ti1 , . . . , tni , f1 (ti1 ), f2 (ti1 , ti2 ), . . . , fn (ti1 , . . . , tni )
Ex `CQC ∃x∀y ϕ(x, y ) ⇔ for some t1 , . . . , tk : `CPC
Wk
i=1
ϕ(ti , f (ti ))
5 / 23
Nonclassical theories
Skolemization in nonclassical logics/theories?
6 / 23
Skolemization for infix formulas
Dfn A quantifier occurrence in ϕ is strong when it is a positive
occurrence of a universal or a negative occurrence of an existential
quantifier. A quantifier occurrence in ϕ is weak if it is not strong.
If put in prenex normal form, strong quantifier occurrences become
universal and weak quantifier occurrences become existential quantifiers.
Dfn The (infix) Skolemization, ϕs , of ϕ is the result of replacing in ϕ all
strong quantifier occurrences
Qxψ(x, ȳ ) by ψ(f (ȳ ), ȳ ),
starting with the leftmost quantifier, where ȳ are the variables of the
weak quantifiers in the scope of which Qxψ(x, ȳ ) occurs and f is a fresh
function symbol in the skolem language Ls ⊇ L.
Thm CQC admits Skolemization:
`CQC ϕ ⇔ `CQC ϕs .
7 / 23
Complexity of Skolemization
Thm (Baaz & Leitsch 1994) Prefix Skolemization may result in a
nonelementary increase of Herbrand complexity (the minimal number of
disjuncts in a Herbrand disjunction).
Thm (Avigad 2003) If the underlying theory allows for a modicum of
coding, Skolem functions can be eliminated in polynomial time.
Thm (Baaz & Hetzl & Weller 2012) The complexity of deSkolemization
is the complexity of the function that given a proof of the Skolemization
of a formula gives the length of the shortest proof of the formula. For
cut-free proofs this complexity is exponential.
Related work on Herbrand expansions in classical logic by Baaz & Hetzl
& Straßburger appeared recently.
8 / 23
Skolemization in nonclassical logics
Note Not all intermediate logics admit Skolemization.
Ex In intuitionistic predicate logic IQC:
DNS
6`IQC ¬¬∃xϕ(x) → ∃x¬¬ϕ(x)
`IQC ¬¬ϕ(c) → ∃x¬¬ϕ(x)
6`IQC ∀x¬¬ϕ(x) → ¬¬∀xϕ(x)
`IQC ∀x¬¬ϕ(x) → ¬¬ϕ(c)
CD 6`IQC ∀x(ϕ(x) ∨ ψ) → ∀xϕ(x) ∨ ψ
`IQC ∀x(ϕ(x) ∨ ψ) → ϕ(c) ∨ ψ.
9 / 23
Two directions
I Develop alternative methods that a given theory admits.
II Establish which theories admit a given alternative method.
Question: What is an alternative Skolemization method?
Dfn Necessary condition for an alternative Skolemization method:
It is a (computable) translation (·)a of formulas in L to formulas in
La ⊇ L such that ϕa does not contain strong quantifiers.
A logic L admits the alternative method if
`L ϕ ⇔ `L ϕa .
The method is strict if for all ϕ and all models K for La :
KL ϕ ⇔ K ϕa ,
where KL the restriction of K to L.
10 / 23
No strict alternatives
Thm (Iemhoff ’16)
No intermediate logic that is sound and complete with respect to a class
of frames, admits a strict, alternative Skolemization method. This holds
for IQC is particular.
Dfn Given a logic L the fragment consisting of formulas without strong
(weak) quantifiers is denoted by Lw (Ls ).
The theorem is a corollary of the following theorem.
Thm For every intermediate logic that is sound and complete with respect
to a class F of frames, the fragments Lw (Ls ) are sound and complete
with respect to the class of constant domain models on frames in F.
11 / 23
No strict alternatives
Thm For every intermediate logic that is sound and complete with respect
to a class F of frames, the fragments Lw (Ls ) are sound and complete
with respect to the class of constant domain models on frames in F.
Dfn Given a rooted Kripke model K , K ↓ is the model obtained by
replacing every domain by the domain at the root, and K ↑ is the model
obtained by replacing every domain by the union of all domains in the
model.
K:
1O R
0N
K↓ :
1O N
K↑ :
0N
1O R
0 R ∀r 6∈ N 6 P(r )
Lem For all formulas ψ without weak quantifiers:
K , k ψ ⇒ K ↓, k ψ
K , k 6 ψ ⇒ K ↑ , k 6 ψ.
For all formulas ψ without strong quantifiers:
K , k ψ ⇒ K ↑, k ψ
K , k 6 ψ ⇒ K ↓ , k 6 ψ.
Cor 6`Lw ϕ implies K ↓ 6 ϕ, and 6`Ls ϕ implies K ↑ 6 ϕ.
12 / 23
Alternatives that are not strict
Dfn IQCE is a conservative extension of IQC with an existence predicate
E such that Et denotes that t exists.
Dfn The eskolemization, ϕe , of ϕ is the result of replacing all strong
quantifier occurrences Qxψ(x, ȳ ) in ϕ (starting with the leftmost) by
(
Ef (ȳ ) ∧ ψ(f (ȳ ), ȳ )
if Q = ∃
Ef (ȳ ) → ψ(f (ȳ ), ȳ ) if Q = ∀,
where ȳ are the variables of the weak quantifiers in the scope of which
Qxψ(x, ȳ ) occurs and f is a fresh function symbol.
Thm (Baaz & Iemhoff ’11)
The logic IQCE, which is a conservative extension of IQC with an
existence predicate E , admits eskolemization for existential quantifiers.
13 / 23
Alternatives that are not strict
Thm (Baaz & Iemhoff ’08)
The logic IQCO, which is a logic with labelled formulas that is a
conservative extension of IQC with an existence predicate and partial
order, admits an alternative Skolemization method called orderization.
Question: In intuitionistic logic Skolem functions are partial?
Related work: Skolemization in fuzzy logics (Baaz, Ciabattoni, Cintula,
Diaconescu, Metcalfe, . . . )
14 / 23
Two directions
I Develop alternative methods that a given logic/theory admits.
II Establish which logics/theories admit the (alternative) Skolemization
method.
Mix: Develop an alternative method and establish which logics/theories
admit the alternative method.
15 / 23
Parallel Skolemization
Dfn (Baaz & Iemhoff) The parallel Skolemization (pskolemization), ϕp ,
of ϕ is the result of replacing, starting with the leftmost, strong
quantifier occurrences Qxψ(x, ȳ ) in ϕ by
( W
n
ψ(fi (ȳ ), ȳ ) if Q = ∃
Wi=1
n
if Q = ∀,
i=1 ψ(fi (ȳ ), ȳ )
where ȳ are the variables of the weak quantifiers in the scope of which
Qxψ(x, ȳ ) occurs, n ∈ N, and the fi are fresh function symbols.
Thm (Baaz & Iemhoff ’14) Every intermediate logic that is sound and
complete with respect to a class of constant domain finite width Kripke
models with the witness property, admits parallel Skolemization.
Cor Every intermediate logic with the constant domain finite model
property admits parallel Skolemization.
Thm (Cintula & Diaconescu & Metcalfe ’15) Every substructural logic
with a semantics that has the n-witnessed model property, admits parallel
Skolemization left and right of degree n.
16 / 23
Parallel Skolemization
Thm (Baaz & Iemhoff ’14) Every intermediate logic that is sound and
complete with respect to a class of constant domain finite width Kripke
models with the witness property, admits parallel Skolemization.
Prf Sufficient: If K 6 ϕ, then there exists a model K p such that K p 6 ϕp .
Simple case: ϕ = ϕ[∃xψ(x)] and ϕp = ϕ[ψ(c1 ) ∨ ψ(c2 )]. To show:
K , k ϕ[∃xψ(x)] ⇔ K p , k ϕ[ψ(c1 ) ∨ ψ(c2 )].
k1 ψ(1)
e
K
k2 ψ(2)
9
k1 ψ(c1 )
e
k0
Kp
k2 ψ(c2 )
9
k0
If c1 is interpreted as 1 and c2 as 2, then for i = 0, 1, 2:
K , ki ∃xψ(x) ⇔ K p , ki ψ(c1 ) ∨ ψ(c2 ).
The witness property garantees the existence of elements in the domain
that can be the interpretation of the ci , or the fi , in the general case. a
17 / 23
Witnesses
Dfn A Kripke model has quantifier witnesses or the witness property if for
every quantified formula Qxψ(x) along every branch there is an element
d such that for all nodes k along that branch:
k Qxψ(x) ⇔ k ψ(d).
..
.O
Ex Constant domain finite
rooted models have quantifier witnesses.
Ex Infinite models in general do
not have quantifier witnesses. A model
such that 0 6 ∀xP(x) without witness:
2O P1
1O P0
Similar notions
exist for t-norm logics and algebras.
0N
(Baaz & Ciabattoni ’16)
Proof theory of witnessed Gödel logic.
18 / 23
Parallel Skolemization without constant domains
Dfn (Baaz & Iemhoff) The epskolemization, ϕp , of ϕ is the result of
replacing, starting with the leftmost, strong quantifier occurrences
Qxψ(x, ȳ ) in ϕ by
( W
n
Efi (ȳ ) ∧ ψ(fi (ȳ ), ȳ )
if Q = ∃
Wi=1
n
if Q = ∀,
i=1 Efi (ȳ ) → ψ(fi (ȳ ), ȳ )
where ȳ are the variables of the weak quantifiers in the scope of which
Qxψ(x, ȳ ) occurs, n ∈ N, and the fi are fresh function symbols.
Thm (Baaz & Iemhoff ’14) Every logic in IQCE that is sound and
complete with respect to a class of finite width Kripke models with the
witness property, admits epskolemization.
Cor Every logic in IQCE with the finite model property admits
epskolemization.
19 / 23
A forgotten (by me) result
In: The eskolemization of universal quantiers, 2010.
Thm Every intermediate logic that is sound and complete with respect to
a class of constant domain Kripke models with the witness property,
admits skolemization.
Prf idea. Simple case: ϕ = ϕ[∃xψ(x)] and ϕs = ϕ[ψ(c)]. To show:
K , ki ∃xψ(x) ⇔ K s , ki ψ(c).
k1 ψ(w1 )
e
K
k0
k2 ψ(w2 )
9
k1 ψ(c)
e
Ks
k2 ψ(c)
9
k0
The domain of K s consists of all closed terms in L ∪ {c} ∪ DK .
¯ for all predicates P.
As long as K , ki 6 ∃xψ(x), K s , ki 6 P(c, d)
¯ ⇔ K , ki P(wi , d).
¯
If K , ki ∃xψ(x), then K s , ki P(c, d)
a
20 / 23
Skolemization without finite width
Thm Every intermediate logic that is sound and complete with respect to
a class of constant domain Kripke models with the witness property,
admits skolemization.
Thm Every logic in IQCE that is sound and complete with respect to a
class of Kripke models with the witness property, admits eskolemization.
Thm The theory iEqe of decidable equality over IQCE admits
eskolemization.
Thm The theory iMPe of decidable monadic predicates over IQCE
admits eskolemization.
21 / 23
Three questions out of many
◦ Does the generalization to logics without finite width extend to
algebraic semantics?
◦ Can the result that there are no strict alternatives to Skolemization
for certain intermediate logics be strengthened?
◦ What do the results imply for theories that do not admit coding,
such as the theory of apartness?
22 / 23
Finis
23 / 23