Logics of Formal Inconsistency enriched with replacement: an algebraic and modal account
aa r X i v : . [ m a t h . L O ] M a r Logics of Formal Inconsistency enriched withreplacement: an algebraic and modal account
Walter Carnielli , Marcelo E. Coniglio and David Fuenmayor Institute of Philosophy and the Humanities - IFCH, andCentre for Logic, Epistemology and the History of Science - CLEUniversity of Campinas, BrazilEmail: { walterac,coniglio } @unicamp.br Freie Universit¨at BerlinBerlin, GermanyEmail: [email protected]
Abstract
One of the most expected properties of a logical system is that it can be algebraiz-able , in the sense that an algebraic counterpart of the deductive machinery could befound. Since the inception of da Costa’s paraconsistent calculi, an algebraic equiv-alent for such systems have been searched. It is known that these systems are nonself-extensional (i.e., they do not satisfy the replacement property). More than this,they are not algebraizable in the sense of Blok-Pigozzi. The same negative results holdfor several systems of the hierarchy of paraconsistent logics known as
Logics of FormalInconsistency ( LFI s). Because of this, these logics are uniquely characterized by se-mantics of non-deterministic kind. This paper offers a solution for two open problemsin the domain of paraconsistency, in particular connected to algebraization of
LFI s, byobtaining several
LFI s weaker than C , each of one is algebraizable in the standardLindenbaum-Tarski’s sense by a suitable variety of Boolean algebras extended with op-erators. This means that such LFI s satisfy the replacement property. The weakest
LFI satisfying replacement presented here is called
RmbC , which is obtained from thebasic
LFI called mbC . Some axiomatic extensions of
RmbC are also studied, and inaddition a neighborhood semantics is defined for such systems. It is shown that
RmbC can be defined within the minimal bimodal non-normal logic E ⊕ E defined by the fu-sion of the non-normal modal logic E with itself. Finally, the framework is extendedto first-order languages. RQmbC , the quantified extension of
RmbC , is shown to besound and complete w.r.t. BALFI semantics. Introduction: The quest for the algebraic counter-part of paraconsistency
One of the most expected properties of a logical system is that it can be algebraizable , inthe sense that an algebraic counterpart of the deductive machinery could be found. Whenthis happens, a lot of logical problems can be faithfully and conservatively translated intosome given algebra, and then algebraic tools can be used to tackle them. This happens sonaturally with the brotherhood between classical logic and Boolean algebra, that a similarrelationship is expected to hold for non-standard logics as well. And indeed it holds for some,but not for all logics. In any case, the task of finding such an algebraic counterpart is farfrom trivial. The intuitive idea behind the search for algebraization for a given logic system,generalizing the pioneering proposal of Lindenbaum and Tarski, usually starts by trying tofind a congruence on the set of formulas that could be used to produce a quotient algebra,defined over the algebra of formulas of the logic.Finding such an algebraization for the logics of the hierarchy C n of da Costa, introducedin [26], constitutes a paradigmatically difficult case. One of the favorite methods to set upcongruences is to check the validity of a fundamental property called replacement or (IpE)(acronym for intersubstitutivity by provable equivalents , intuitively clear, and to be formallydefined in Section 2. A logic enjoying replacement is usually called self-extensional .It is known since some time that (IpE) does not hold for C , the first logic of da Costa’sfamily. A proof can be found in [21] (Corollary 3.65); as a consequence, a direct Lindenbaum-Tarski algebraization for this logic is not possible. This closes the way to the other, weakercalculi of the hiearchy C n , since when one logic is algebraizable, so are its extensions. Butthere are other possibilities for algebraization, and the search continued until a proof waspresented by Mortensen in 1980 [34], establishing that no non-trivial quotient algebra isdefinable for C , or for any logic weaker than C . In 1991, an even more negative result,found by Lewin, Mikenberg, Schwarze (see [31]) shows that C is not even algebraizablein the more general sense of Blok-Pigozzi (see [9]). This result was generalized in [21,Theorem 3.83] to Cila , the presentation of C in the language of the Logics of FormalInconsistency ( LFI s) featuring a (primitive) consistency conective ◦ . Since any deductiveextension of an algebraizable logic (in the same language) is also algebraizable, we obtain asa consequence that no such algebraization is possible for any other of the LFI s weaker than
Cila studied in [21, 17, 14], like mbC , mbCciw , bC and Ci . The same reasoning appliesto every calculus C n in the infinite da Costa’s hierarchy, given that they are weaker than C .Some extensions of C having non-trivial quotient algebras have been proposed in theliterature. In [35], for instance, Mortensen has proposed an infinite number of intermediatelogics between C and classical logic called C n/ ( n +1) , for n ≥
1. Such logics were shown toenjoy non-trivial congruences defined by finite sets of equations for each n ≥
1, being thusalgebraizable in the sense of Blok-Pigozzi (though not in the traditional sense of Lindenbaum-Tarski).Some other types of algebraic counterparts have been investigated, for instance, in [19]and [39] an algebraic variety (da Costa algebras) for the logic C was defined, permitting aStone-like representation theorem. In this way, every da Costa algebra is isomorphic to aparaconsistent algebra of sets, making C closer to traditional mathematical objects.2t can be proved, however, that for some subclasses of LFI s such intersubstitutivity resultsis unattainable, as shown in Theorem 3.51 of [21] with respect to the logic Ci , one of thecentral systems of the family of LFI s which is much weaker than
Cila .Some interesting results concerning three-valued self-extensional paraconsistent logicswere obtained in the literature, in connection with the limitative result [21, Theorem 3.51]mentioned above. In [3] it was shown that no three-valued paraconsistent logic having animplication can be self-extensional. On the other hand, in [2] it was shown that there isexactly one self-extensional three-valued paraconsistent logic defined in a signature havingconjunction, disjunction and negation. For paraconsistent logics in general, it was shownin [6] that no paraconsistent negation ¬ satisfying the law of double negation and such thatthe schema ¬ ( ϕ ∧ ¬ ϕ ) is valid can satisfy (IpE).Nevertheless, there was still an open question: to obtain (IpE) for extensions of Ci bythe addition of weaker forms of contraposition deduction rules, as discussed in Subsection3.7 of [21]. The challenge was to find extensions of bC and Ci which would satisfy (IpE)and still keep their paraconsistent character. In this paper we meet this challenge. We definethe logic RmbC , an extension by rules of mbC , and two suitable extensions of
RmbC , thelogics
RbC and
RCi (respectively, extensions of bC and Ci ) that solve the open problem.Details are given in Example 3.9 of Section 3. A new kind of semantic structures, the Boolean algebras with
LFI operators, or BALFIs,a generalization of BAOs (Boolean algebras with operators) is introduced in Section 2, and
RmbC is proved to be sound and complete w.r.t. BALFIs.The paper also investigates some other directions. Section 4 studies the limits for replace-ment under the conditions for paraconsistency, and Section 5 proposes neigborhood semanticsfor
RmbC as a special class of BALFIs defined on powerset Boolean algebras. Again,
RmbC is proved to be sound and complete w.r.t. such version of neigborhood models. Moreover, inSection 6 it is proved that
RmbC can be defined within the minimal bimodal non-normalmodal logic. This neigborhood semantics is also proposed for axiomatic extensions of
RmbC in Section 7.A special problem is studied in Section 8: the BALFI semantics for
RmbC , as well as itsneigborhood semantics defined in Section 5, are degree-preserving instead of truth-preserving(in the sense of [10]). This requires adapting the usual definition of derivation from premisesin a Hilbert calculus (cf. Definition 2.6). But it is also possible to consider global (or truth-preserving) semantics, as it is usually done with algebraic semantics. This leads us to thelogic
RmbC ∗ , which is defined by the same Hilbert calculus than the one for RmbC , butwhere derivations from premises are defined as in the usual Hilbert calculi.Section 9 is dedicated to extending
RmbC to first-order languages, defining the logic
RQmbC , which is proved, in Section 10 and Section 11, to be complete w.r.t. BALFIsemantics. The proof is an adaptation to the completeness proof for
QmbC w.r.t. swapstructures semantics given in [24], and since BALFIs are ordinary algebras, the new com-pleteness proof offers a great simplification when compared to previous completeness resultsbased on non-deterministic swap structures. To generate heuristics and suitable models, as well as to block dead-ends by finding counter-models, wecount with the help of the proof assistant Isabelle/HOL. The logic RmbC
The class of paraconsistent logics known as
Logics of Formal Inconsistency ( LFI s, for short)was introduced by W. Carnielli and J. Marcos in [21]. In their simplest form, they have anon-explosive negation ¬ , as well as a (primitive or derived) consistency connective ◦ whichallows to recover the Law of Explosion in a controlled way. Definition 2.1
Let L = h Θ , ⊢i be a Tarskian, finitary and structural logic defined over apropositional signature Θ , which contains a negation ¬ , and let ◦ be a (primitive or defined)unary connective. Then, L is said to be a Logic of Formal Inconsistency with respect to ¬ and ◦ if the following holds:(i) ϕ, ¬ ϕ ψ for some ϕ and ψ ;(ii) there are two formulas α and β such that(ii.a) ◦ α, α β ;(ii.b) ◦ α, ¬ α β ;(iii) ◦ ϕ, ϕ, ¬ ϕ ⊢ ψ for every ϕ and ψ . Condition (ii) of the definition of
LFI s is required in order to satisfy condition (iii) ina non-trivial way. The hierarchy of
LFI s studied in [17] and [14] starts from a logic called mbC , which extends positive classical logic
CPL + by adding a negation ¬ and a unary consistency operator ◦ satisfying minimal requirements in order to define an LFI . Definition 2.2
From now on, the following signatures will be considered: Σ + = {∧ , ∨ , →} ; Σ BA = {∧ , ∨ , → , ¯0 , ¯1 } ; Σ = {∧ , ∨ , → , ¬ , ◦} ; Σ C = {∧ , ∨ , → , ¬} ; Σ C = {∧ , ∨ , → , ¬ , ¯0 } ; Σ C e = {∧ , ∨ , → , ¬ , ¯0 , ¯1 } ; Σ e = {∧ , ∨ , → , ¬ , ◦ , ¯0 , ¯1 } ; Σ m = {∧ , ∨ , → , ∼ , (cid:3) , ♦ } ; and Σ bm = {∧ , ∨ , → , ∼ , (cid:3) , ♦ , (cid:3) , ♦ } . If Θ is a propositional signature, then
F or (Θ) will denote the (absolutely free) algebra offormulas over Θ generated by a given denumerable set V = { p n : n ∈ N } of propositionalvariables. 4 efinition 2.3 (Classical Positive Logic) The classical positive logic
CPL + is definedover the language F or (Σ + ) by the following Hilbert calculus: Axiom schemas: α → (cid:0) β → α (cid:1) ( Ax1 ) (cid:16) α → (cid:0) β → γ (cid:1)(cid:17) → (cid:16)(cid:0) α → β (cid:1) → (cid:0) α → γ (cid:1)(cid:17) ( Ax2 ) α → (cid:16) β → (cid:0) α ∧ β (cid:1)(cid:17) ( Ax3 ) (cid:0) α ∧ β (cid:1) → α ( Ax4 ) (cid:0) α ∧ β (cid:1) → β ( Ax5 ) α → (cid:0) α ∨ β (cid:1) ( Ax6 ) β → (cid:0) α ∨ β (cid:1) ( Ax7 ) (cid:16) α → γ (cid:17) → (cid:16) ( β → γ ) → (cid:0) ( α ∨ β ) → γ (cid:1)(cid:17) ( Ax8 ) (cid:0) α → β (cid:1) ∨ α ( Ax9 ) Inference rule: α α → ββ ( MP ) Definition 2.4
The logic mbC , defined over signature Σ , is obtained from CPL + by addingthe following axiom schemas: α ∨ ¬ α ( Ax10 ) ◦ α → (cid:16) α → (cid:0) ¬ α → β (cid:1)(cid:17) ( bc1 )The logic mbC is an LFI . Indeed, it is the minimal
LFI extending
CPL + .Consider the Replacement property, namely: If α ↔ β is a theorem then γ [ p/α ] ↔ γ [ p/β ]is a theorem, for every formula γ ( p ) (as usual, α ↔ β is an abbreviation of the formula ( α → β ) ∧ ( β → α ), and γ [ p/α ] denotes the formula obtained from γ by replacing every occurrenceof the variable p by the formula α ). It is well known that mbC does not satisfy replacementin general. However, it is easy to see that replacement holds in mbC for every formula γ ( p )over the signature Σ + of CPL + . We introduce now the logic RmbC which extends mbC by adding replacement for every formula over Σ. From the previous observation, it is enoughto add replacement for ¬ and ◦ as new inference rules. Namely: if α ↔ β is a theorem then ¬ α ↔ ¬ β (is a theorem), and if α ↔ β is a theorem then ◦ α ↔ ◦ β (is a theorem).Observe, however, that replacement is in fact a metaproperty (since it states that someformula is a theorem from previous formulas which are assumed to be theorems). It is clearthat the two inference rules proposed above for inducing replacement are global instead oflocal (see Section 8 below): in order to apply each rule, the corresponding premise must be atheorem. This is an analogous situation to the Necessitation rule in modal logics. Assuminginference rules of this kind requires changing the definition of derivation from premises in theresulting Hilbert calculus, as we shall see below.5 efinition 2.5
The logic
RmbC , defined over signature Σ , is obtained from mbC by addingthe following inference rules: α ↔ β ¬ α ↔ ¬ β ( R ¬ ) α ↔ β ◦ α ↔ ◦ β ( R ◦ ) Definition 2.6 (Derivations in RmbC) (1) A derivation of a formula ϕ in RmbC is a finite sequence of formulas ϕ . . . ϕ n suchthat ϕ n is ϕ and, for every ≤ i ≤ n , either ϕ i is an instance of an axiom of RmbC , or ϕ i is the consequence of some inference rule of RmbC whose premises appear in the sequence ϕ . . . ϕ i − .(2) We say that a formula ϕ is derivable in RmbC , or that ϕ is a theorem of RmbC ,denoted by ⊢ RmbC ϕ , if there exists a derivation of ϕ in RmbC .(3) Let Γ ∪{ ϕ } be a set of formulas over Σ . We say that ϕ is derivable in RmbC from Γ , andwe write Γ ⊢ RmbC ϕ , if either ϕ is derivable in RmbC , or there exists a finite, non-emptysubset { γ , . . . , γ n } of Γ such that the formula ( γ ∧ ( γ ∧ ( . . . ∧ ( γ n − ∧ γ n ) . . . ))) → ϕ isderivable in RmbC . Remarks 2.7 (1) From the previous definition, it follows that ∅ ⊢
RmbC ϕ iff ⊢ RmbC ϕ .(2) Recall that a consequence relation ⊢ is said to be Tarskian and finitary if it satisfies thefollowing properties: (i) Γ ⊢ α whenever α ∈ Γ ; (ii) if Γ ⊢ α and Γ ⊆ ∆ then ∆ ⊢ α ;(iii) if Γ ⊢ ∆ and ∆ ⊢ α then Γ ⊢ α , where Γ ⊢ ∆ means that Γ ⊢ δ for every δ ∈ ∆ ;and (iv) Γ ⊢ α implies that Γ ⊢ α for some finite Γ contained in Γ . It can be proventhat the consequence relation ⊢ RmbC given in Definition 2.6(2) is Tarskian and finitary, byusing a general result stated by W´ojcicki in [40]. Specifically, in Section 2.10 of that bookit was studied the question of characterizing a Tarskian consequence relation ⊢ in termsof theoremhood, provided that the language contains an implication ⇒ and a conjunction & . Namely, the problem is to find necessary and sufficient conditions in order to have that γ , . . . , γ n ⊢ ϕ iff ⊢ ( γ & ( γ & ( . . . & ( γ n − & γ n ) . . . ))) ⇒ ϕ and still having that ⊢ isTarskian and finitary. Thus, in item (ii) of Theorem 2.10.2 in [40] certain requirementswere found for ⇒ and & which are necessary and sufficient to guarantee that a consequencerelation defined as in Definition 2.6 is Tarskian and finitary. It is easy to prove, by usingthe properties of CPL + , that → and ∧ satisfy such requirements in RmbC . From this, itfollows that
RmbC is indeed a Tarskian and finitary logic.
By the properties of ∧ and → inherited from CPL + , and by the notion of derivation in RmbC , it is easy to see that the
Deduction Metatheorem holds in
RmbC : Theorem 2.8 (Deduction Metatheorem for RmbC) Γ , ϕ ⊢ RmbC ψ if and only if Γ ⊢ RmbC ϕ → ψ . The problem was originally presented in [40] in a more general way. We are presenting here a particularcase of that problem, which is enough to our purposes. Moreover, in [40] the problem was analyzed interms of Tarskian consequence operators instead of Tarskian consequence relations, but both formalisms areequivalent in this context. Of course the satisfaction of the Deduction Theorem is what lies behind the problem studied in [40]mentioned in Remark 2.7(2).
RmbC will be given by means of a suitable class of Boolean algebraswith additional operators. Because of the definition of deductions in
RmbC discussedabove, the semantic consequence relation will be degree preserving instead of truth preserving (see [10]). In modal terms, the semantics will be local instead of global. We will return tothis point in Section 8.
Definition 2.9 (BALFIs) A Boolean algebra with
LFI operators (BALFI, for short) is analgebra B = h A, ∧ , ∨ , → , ¬ , ◦ , , i over Σ e such that its reduct A = h A, ∧ , ∨ , → , , i to Σ BA is a Boolean algebra and the unary operators ¬ and ◦ satisfy: a ∨ ¬ a = 1 and a ∧ ¬ a ∧ ◦ a = 0 ,for every a ∈ A . The variety of BALFIs will be denoted by BI . Definition 2.10 (Degree-preserving BALFI semantics) (1) A valuation over a BALFI B is a homomorphism v : F or (Σ) → B .(2) Let ϕ be a formula in F or (Θ) . We say that ϕ is valid in BI , denoted by | = BI ϕ , if, forevery BALFI B and every valuation v over it, v ( ϕ ) = 1 .(3) Let Γ ∪{ ϕ } be a set of formulas in F or (Θ) . We say that ϕ is a local (or degree-preserving )consequence of Γ in BI , denoted by Γ | = BI ϕ , if either ϕ is valid in BI , or there exists a finite,non-empty subset { γ , . . . , γ n } of Γ such that, for every BALFI B and every valuation v overit, V ni =1 v ( γ i ) ≤ v ( ϕ ) . Remark 2.11
Note that Γ | = BI ϕ iff either ϕ is valid in BI , or there exists a finite, non-empty subset { γ , . . . , γ n } of Γ such that ( γ ∧ ( γ ∧ ( . . . ∧ ( γ n − ∧ γ n ) . . . ))) → ϕ is valid. Thisfollows easily from the definitions, and from the fact that a ≤ b iff a → b = 1 in any Booleanalgebra A . Theorem 2.12 (Soundness of RmbC w.r.t. BI ) Let Γ ∪ { ϕ } ⊆ F or (Θ) . Then: Γ ⊢ RmbC ϕ implies that Γ | = BI ϕ . Proof.
Let ϕ be a an instance of an axiom of RmbC . It is immediate to see that, for every B and every valuation v on it, v ( ϕ ) = 1. Now, let α, β ∈ F or (Σ). If v ( α → β ) = 1 and v ( α ) = 1 then, since v ( α → β ) = v ( α ) → v ( β ), it follows that v ( β ) = 1. On the otherhand, if v ( α ↔ β ) = 1 then v ( α ) = v ( β ) and so v ( α ) = v ( α ) = v ( β ) = v ( β ), henceit follows that v ( α ↔ β ) = 1 for ∈ {¬ , ◦} . From this, by induction on the lengthof a derivation of ϕ in RmbC , it can be easily proven that ϕ is valid in BI whenever ϕ isderivable in RmbC . Now, suppose that Γ ⊢ RmbC ϕ . If ⊢ RmbC ϕ then, by the observationabove, ϕ is valid in BI and so Γ | = BI ϕ . On the other hand, if there exists a finite, non-emptysubset { γ , . . . , γ n } of Γ such that ⊢ RmbC ( γ ∧ ( γ ∧ ( . . . ∧ ( γ n − ∧ γ n ) . . . ))) → ϕ then, onceagain by the observation above, | = BI ( γ ∧ ( γ ∧ ( . . . ∧ ( γ n − ∧ γ n ) . . . ))) → ϕ . This shows thatΓ | = BI ϕ , by Remark 2.11. ✷ Theorem 2.13 (Completeness of RmbC w.r.t. BI ) Let Γ ∪ { ϕ } ⊆ F or (Θ) . Then: Γ | = BI ϕ implies that Γ ⊢ RmbC ϕ . It is worth noting that these operators not necessarily commute with joins. Thus, the algebras are notnecessarily coincident with the so-called
Boolean algebras with operators (BAOs) used as semantics for modallogics (see, for instance, [30]). roof. Define the following relation on
F or (Σ): α ≡ β iff ⊢ RmbC α ↔ β . It is clearly anequivalence relation, by the properties of CPL + . Let A can def = F or (Σ) / ≡ be the quotient set,and define over A can the following operations: [ α ] β ] def = [ α β ], for ∈ {∧ , ∨ , →} , where[ α ] denotes the equivalence class of α w.r.t. ≡ . Let 0 def = [ α ∧ ¬ α ∧ ◦ α ] and 1 def = [ α ∨ ¬ α ].These operations and constants are clearly well-defined, and so they induce a structure ofBoolean algebra over the set A can , which will be denoted by A can . Let B can be its expansionto Σ e by defining α ] def = [ α ], for ∈ {¬ , ◦} . These operations are well-defined, and itis immediate to see that B can is a BALFI. Let v can : F or (Σ) → B can given by v can ( α ) = [ α ].Clearly v can is a valuation over B can such that v can ( α ) = 1 iff ⊢ RmbC α .Now, suppose that Γ | = BI ϕ , and recall Remark 2.11. If | = BI ϕ then, in particular, v can ( ϕ ) = 1 and so ⊢ RmbC ϕ . From this, Γ ⊢ RmbC ϕ . On the other hand, if there exists afinite, non-empty subset { γ , . . . , γ n } of Γ such that | = BI ( γ ∧ ( γ ∧ ( . . . ∧ ( γ n − ∧ γ n ) . . . ))) → ϕ then, in particular, v can (( γ ∧ ( γ ∧ ( . . . ∧ ( γ n − ∧ γ n ) . . . ))) → ϕ ) = 1. This means that ⊢ RmbC ( γ ∧ ( γ ∧ ( . . . ∧ ( γ n − ∧ γ n ) . . . ))) → ϕ and so Γ ⊢ RmbC ϕ . ✷ Definition 2.14
The pair hB can , v can i defined in the proof of Theorem 2.13 is called thecanonical model of RmbC . Example 2.15 (BALFIs over ℘ ( { w , w } ) ) Let A = ℘ ( { w , w } ) = { , a, b, } be thepowerset of W = { w , w } such that ∅ , a = { w } , b = { w } and { w , w } .Then, the BALFIs defined by expanding the Boolean algebra A are the following (below, | separates the possible options for the values of ¬ z and ◦ z for every value of z , while x standsfor any element of A ): z ¬ z ◦ z | a | b | x | (0 or b ) | (0 or a ) | a b | x | (0 or b ) b a | x | (0 or a )0 1 x On each row, each choice in the ith position of the sequence of options in the column for ¬ z forces a choice in the ith position of the sequence of options in the column for ◦ z . Forinstance, if in the current BALFI we choose ¬ b then there are two possibilities for thevalue of ◦ in that BALFI: either ◦ or ◦ a . On the other hand, by choosing that ¬ a = 1 it forces that either ◦ a = 0 or ◦ a = b . Otherwise, if ¬ a = b then ◦ a can be arbitrarilychosen. Remark 2.16
Observe that the rules ( R ¬ ) and ( R ◦ ) do not ensure that ⊢ RmbC ( α ↔ β ) → ( ¬ α ↔ ¬ β ) and ⊢ RmbC ( α ↔ β ) → ( ◦ α ↔ ◦ β ) in general. Consider, for instance α = p and β = q where p and q are two different propositional variables, and take the following BALFI B defined over the Boolean algebra ℘ ( { w , w } ) , according to Example 2.15: z ¬ z ◦ z a b ab a b ow, consider a valuation v over B such that v ( p ) = a and v ( q ) = 1 . Hence v ( ¬ p ) = b , v ( ¬ q ) = 1 , v ( ◦ p ) = a and v ( ◦ q ) = 0 . From this v ( p ↔ q ) = a and v ( ¬ p ↔ ¬ q ) = v ( ◦ p ↔◦ q ) = b . Therefore v (( p ↔ q ) → ( ¬ p ↔ ¬ q )) = v (( p ↔ q ) → ( ◦ p ↔ ◦ q )) = b . That is, noneof the last two formulas is valid in RmbC . Of course both formulas hold if ⊢ RmbC ( α ↔ β ) ,by ( R ¬ ) and ( R ◦ ).Clearly RmbC is an
LFI : in the BALFI B we just defined above, the given valuation v shows that q, ¬ q RmbC p . Now, consider the following BALFI B ′ defined over ℘ ( { w , w } ) ,using again Example 2.15: z ¬ z ◦ z a b Take a valuation v ′ over B ′ such that v ′ ( p ) = 1 and v ′ ( q ) = a . This shows that p, ◦ p RmbC q .Now, a valuation v ′′ over B ′ such that v ′′ ( p ) = 0 and v ′′ ( q ) = b shows that ¬ p, ◦ p RmbC q . Inaddition, a valuation v ′′′ over B ′ such that v ′′′ ( p ) = a and v ′′′ ( q ) = b shows that p, ¬ p RmbC q .On the other hand, by Definition 2.9 it is the case that α, ¬ α, ◦ α ⊢ RmbC β for every formulas α and β . In [21], the first study on
LFI s, the replacement property was analyzed under the name(IpE), presented in the following (equivalent) way:(IpE) if α i ⊣⊢ β i (for 1 ≤ i ≤ n ) then ϕ ( α , . . . , α n ) ⊣⊢ ϕ ( β , . . . , β n )for every formulas α i , β i , ϕ . In that article, an important question was posed: to find ex-tensions of bC and Ci (two axiomatic extensions of mbC to be analyzed below) whichsatisfy (IpE) still being paraconsistent. In this section, we will show a solution to that openproblem, obtained by extending axiomatically
RmbC . In what follows, some
LFI s whichare axiomatic extensions of mbC ( bC and Ci , among others) will be considered, and themethodology adopted for RmbC to such extensions will be adapted in a suitable way. In [21], page 41, we can read: “The question then would be if (IpE) could be obtained for real
LFI s”.On page 54, after observing that in extensions of Ci it is enough ensuring (IpE) for negation, since it implies(IpE) for ◦ , it is said that “We suspect that this can be done, but we shall leave it as an open problem atthis point”. Finally, they observe on page 55, footnote 17 that certain 8-valued matrices presented by Urbassatisfy (IpE) for an extension of bC . However, this logic is not paraconsistent. After this, they claim that“the question is still left open as to whether there are paraconsistent such extensions of bC !”. efinition 3.1 (Some extensions of mbC) Consider the following axioms presented in [21]and [14]: ◦ α ∨ ( α ∧ ¬ α ) ( ciw ) ¬◦ α → ( α ∧ ¬ α ) ( ci ) ¬ ( α ∧ ¬ α ) → ◦ α ( cl ) ¬¬ α → α ( cf ) α → ¬¬ α ( ce )( ◦ α ∧ ◦ β ) → ◦ ( α ∧ β ) ( ca ∧ )( ◦ α ∧ ◦ β ) → ◦ ( α ∨ β ) ( ca ∨ )( ◦ α ∧ ◦ β ) → ◦ ( α → β ) ( ca → ) Definition 3.2
Let B = h A, ∧ , ∨ , → , ¬ , ◦ , , i be a BALFI, and let ϕ be a formula over Σ . We say that B is a model of ϕ (considered as an axiom schema), denoted by B (cid:13) ϕ , if v ( σ ( ϕ )) = 1 for every substitution for variables σ : V →
F or (Σ) and every valuation v over B . The proof of the following result is immediate from the definitions:
Proposition 3.3
Let B = h A, ∧ , ∨ , → , ¬ , ◦ , , i be a BALFI. Then:(1) B is a model of ( ciw ) iff B satisfies the equation ◦ a = ∼ ( a ∧ ¬ a ) for every a ∈ A ;(2) B is a model of ( ci ) iff B satisfies the equation ¬◦ a = a ∧ ¬ a for every a ∈ A ;(3) B is a model of ( cl ) iff B satisfies the equation ◦ a = ¬ ( a ∧ ¬ a ) for every a ∈ A ;(4) B is a model of ( cf ) iff B satisfies the equation a ∧ ¬¬ a = ¬¬ a for every a ∈ A ;(5) B is a model of ( ce ) iff B satisfies the equation a ∧ ¬¬ a = a for every a ∈ A ;(6) B is a model of ( ca ) iff B satisfies the equation ( ◦ a ∧ ◦ b ) ∧ ◦ ( a b ) = ◦ a ∧ ◦ b for every a, b ∈ A , for each ∈ {∧ , ∨ , →} . Let Ax be a set formed by one or more of the axiom schemas introduced in Definition 3.1,and let mbC ( Ax ) be the logic defined by the Hilbert calculus obtained from mbC by addingthe set Ax of axiom schemas. Let BI ( Ax ) be the class of BALFIs which are models of everyaxiom in Ax . Clearly, BI ( Ax ) is a variety of algebras. Finally, let RmbC ( Ax ) be the logicobtained from RmbC by adding the set Ax of axiom schemas. It is simple to adapt theproofs of Theorems 2.12 and 2.13 (in particular, by defining for each logic the correspondingcanonical model, as in Definition 2.14) in order to obtain the following: Theorem 3.4 (Soundness and completeness of RmbC ( Ax ) w.r.t. BI ( Ax ) ) Let Γ ∪ { ϕ } ⊆ F or (Σ) . Then: Γ ⊢ RmbC ( Ax ) ϕ if and only if Γ | = BI ( Ax ) ϕ . From this important result, some properties of the calculi
RmbC ( Ax ) can be easilyproven by algebraic methods, that is, by means of BALFIs. For instance: Axiom ( ciw ) was introduced by Avron in [1] by means of two axioms, ( k1 ): ◦ α ∨ α and ( k2 ): ◦ α ∨ ¬ α .Strictly speaking, ( k1 ) and ( k2 ) were presented as rules in a standard Gentzen calculus. roposition 3.5 BI ( { ci , cf } ) = BI ( { cl , cf } ) = BI ( { ci , cl , cf } ) . Hence, the logics RmbC ( { ci , cf } ) , RmbC ( { cl , cf } ) and RmbC ( { ci , cl , cf } ) coincide. Proof. (1) Since ⊢ mbC ( α ∧ ¬ α ) → ¬◦ α and ⊢ mbC ◦ α → ¬ ( α ∧ ¬ α ) then, for every BALFI B and every a ∈ A , ( a ∧ ¬ a ) ≤ ¬◦ a and ◦ a ≤ ¬ ( a ∧ ¬ a ). Let B ∈ BI ( { ci , cf } ), and let a ∈ A . Then a ∧ ¬ a = ¬◦ a and so ¬ ( a ∧ ¬ a ) = ¬¬◦ a ≤ ◦ a . Therefore B ∈ BI ( { cl , cf } ).Conversely, suppose that B ∈ BI ( { cl , cf } ) and let a ∈ A . Since ◦ a = ¬ ( a ∧ ¬ a ) then ¬◦ a = ¬¬ ( a ∧ ¬ a ) ≤ ( a ∧ ¬ a ). From this, B ∈ BI ( { ci , cf } ). This shows the first part of theProposition. The second part follows from Theorem 3.4. ✷ Example 3.6 (BALFIs for RmbCciw)
The logic mbC ( ciw ) was considered in [14] un-der the name mbCciw . This logic was introduced in [1] under the name B [ { ( k1 ) , ( k2 ) } ] ,presented by means of a standard Gentzen calculus such that B is a Gentzen calculus for mbC . The logic mbCciw is the least extension of mbC in which the consistency connectivecan be defined in terms of the other connectives, namely: ◦ α is equivalent to ∼ ( α ∧ ¬ α ) ,where ∼ denotes the classical negation definable in mbC as ∼ α = α → ⊥ . Here, ⊥ denotesany formula of the form β ∧ ¬ β ∧ ◦ β . Let
RmbCciw be the logic
RmbC ( ciw ). Becauseof the satisfaction of the replacement property, and given that the consistency connective canbe defined in terms of the other connectives, the connective ◦ can be eliminated from thesignature, and so we consider the logic RmbCciw defined over the signature Σ C (recallDefinition 2.2), obtained from CPL + by adding ( Ax10 ) , ( R ¬ ) , and axiom schema ( Bot ) : ¯0 → α . In such presentation of RmbCciw , the strong negation is defined by the formula ∼ α = α → ¯0 . The algebraic models for this presentation of RmbCciw are given by BALFIs B = h A, ∧ , ∨ , → , ¬ , , i over Σ C e such that its reduct A = h A, ∧ , ∨ , → , , i to Σ BA is aBoolean algebra and the unary operator ¬ satisfies a ∨ ¬ a = 1 for every a ∈ A . On the otherhand, the expression ◦ a is an abbreviation for ∼ ( a ∧ ¬ a ) in such BALFIs. It is also interesting to observe that ◦ satisfies a sort of necessitation rule in certainextensions of RmbC : Proposition 3.7
Consider the
Necessitation rule for ◦ : α ◦ α ( N ec ◦ ) Then, ( N ec ◦ ) is an admissible rule in RmbC ( { cl , ce } ) . Proof.
Assume that ⊢ RmbC ( { cl , ce } ) α . By the rules of CPL + it follows that ⊢ RmbC ( { cl , ce } ) β ↔ ( α ∧ β ) for every formula β . In particular, ⊢ RmbC ( { cl , ce } ) ¬ α ↔ ( α ∧ ¬ α ) and so, by( R ¬ ), ⊢ RmbC ( { cl , ce } ) ¬¬ α ↔ ¬ ( α ∧ ¬ α ). On the other hand, from ⊢ RmbC ( { cl , ce } ) α it follows Rigorously speaking, ◦ is not defined in terms of the other connectives, since ◦ is essential on order todefine ⊥ . So, the right signature for mbCciw and its extensions is Σ C . Recall that a structural inference rule is admissible in a logic L if the following holds: whenever thepremises of an instance of the rule are theorems of L , then the conclusion of the same instance of the rule isa theorem of L . ⊢ RmbC ( { cl , ce } ) ¬¬ α , by ( ce ) and ( MP ). Then ⊢ RmbC ( { cl , ce } ) ¬ ( α ∧ ¬ α ). By ( cl ) and( MP ) we conclude that ⊢ RmbC ( { cl , ce } ) ◦ α . ✷ Now, we can provide a solution to the first open problem posed in [21]:
Example 3.8 (A paraconsistent extension of bC with replacement)
Consider the logic bC introduced in [21]. By using the notation proposed above, bC corresponds to mbC ( cf ). Then
RbC (that is,
RmbC ( cf )) is an extension of bC which satisfies replacement while itis still paraconsistent. Moreover, RbC is an
LFI . These facts can be easily proven by usingthe BALFI B ′ considered in Remark 2.16. In fact, it is immediate to see that B ′ is a modelof ( cf ) . It is worth noting that B ′ is not a model of ( ciw ) : ◦ a = ∼ ( a ∧ ¬ a ) = ∼ a = b .Therefore, B ′ is neither a model of ( ci ) nor of ( cl ) , given that any of these axioms implies ( ciw ) . We can now offer a solution to the second open problem posed in [21]:
Example 3.9 (A paraconsistent extension of Ci with replacement)
Now, consider thelogic Ci introduced in [21], which corresponds to mbC ( { cf , ci } ), and let RCi = RmbC ( { cf , ci } ). By Proposition 3.5, RCi also derives the schema ( cl ) . It can be proven that RCi is anextension of Ci which satisfies replacement while it is still paraconsistent. In order to provethis, consider the following BALFI B ′′ defined over the Boolean algebra A = ℘ ( { W } ) , thepowerset of W = { w , w , w , w } (note that ∅ and W ): z ¬ z ◦ z { w , w } { w , w , w } { w , w , w }{ w , w } { w , w , w } { w , w , w } X W \ X W where X is different to { w , w } and { w , w } . It is immediate to see that B ′′ is a BALFIfor RCi . Hence, using this model it follows easily that
RCi is a paraconsistent extension of Ci which satisfies (IpE) and ( cl ) . Another paraconsistent model for RCi defined over A isthe following: z ¬ z ◦ z { w , w } { w , w , w } { w , w , w }{ w , w } { w , w , w } { w , w , w }{ w , w } { w , w , w } { w , w , w }{ w , w } { w , w , w } { w , w , w }{ w , w } { w , w , w } { w , w , w }{ w , w } { w , w , w } { w , w , w } X W \ X W We will write mbC ( ϕ ), RmbC ( ϕ ) and BI ( ϕ ) instead of mbC ( { ϕ } ), RmbC ( { ϕ } ) and BI ( { ϕ } ), respec-tively. here the cardinal of X is different to . Example 3.10
We can offer now a model of
RmbC ( cl ) which does not satisfy axiom ( cf ) .Thus, consider the following BALFI B ′′′ defined over the Boolean algebra A = ℘ ( { w , w } ) = { , a, b, } according to Example 2.15: z ¬ z ◦ z a ba b b a
10 1 1
Observe that B ′′′ (cid:13) cl . However, B ′′′ is not a model of ( cf ) : ¬¬ a . In [21, Theorem 3.51] some sufficient conditions were given to show that certain extensionsof bC and Ci cannot satisfy replacement while being still paraconsistent. This result showsthat there are limits, much before reaching classical logic CPL , for extending
RmbC whilepreserving paraconsistency. This result will be applied now in order to give two importantexamples of
LFI s which cannot be extended with replacement to the price of losing paracon-sistency.The first example to be given is, in fact, a family of 8,192 examples:
Example 4.1 (Three-valued LFIs)
Recall the family of 8Kb three-valued
LFI s introduced by Marcos in an unpublished draft, anddiscussed in [21, Section 3.11] and in [17, Section 5.3]. As it was observed in these references,the schema ¬ ( α ∧ ¬ α ) is valid in all of these logics. In addition, all these logics are models ofaxioms ( ci ) and ( cf ) (see, for instance, [17, Theorem 130]). But in [21, Theorem 3.51(ii)]it was proved that ( IpE ) cannot hold in any paraconsistent extension of Ci in which theschema ¬ ( α ∧ ¬ α ) is valid. As a consequence of this, the inference rules ( R ¬ ) and ( R ◦ ) cannot be added to any of them to the price of losing paraconsistency. Indeed, if L is anyof such three-valued logics, the corresponding logic RL obtained by adding both rules willderive the axiom schema ◦ α (the proof of this fact can be easily adapted from the one forTheorem 3.51(ii) presented in [21]). But then, the negation ¬ is explosive in RL , by ( bc1 ) and ( MP ) . This shows that these three-valued LFI s, including the well-known da Costa andD’Ottaviano’s logic J3 (and so all of its equivalent presentations, such as LFI1 , CLuN or LPT0 ), as well as Sette’s logic P1 , if extended by the inference rules proposed here, are nolonger paraconsistent. Of course this result is related to the one obtained in [3], which statesthat for no three-valued paraconsistent logic with implication the replacement property canhold. It is worth noting that with the help of the model finder
Nitpick , which is part of the automated toolsintegrated into Isabelle/HOL [36], we carried out many of the experiments leading to the generation of thetwo models presented here. C , introduced by da Costa in 1963. Example 4.2 (da Costa’s logic C ) In [26] da Costa introduced his famous hierarchy of paraconsistent systems C n (for n ≥ ),the first systematic study on paraconsistency introduced in the literature. As discussed above,da Costa’s approach was generalized through the notion of LFI s. The first and strongersystem in the hierarchy is C , which is equivalent (up to language) with Cila . The logic
Cila corresponds, with the notation introduced above, to mbC ( { ci , cl , cf , ca ∧ , ca ∨ , ca → } ) . If weconsider now RmbC ( { ci , cl , cf , ca ∧ , ca ∨ , ca → } ) , which will be called RCila , then this logicderives ( ciw ) . Indeed, as shown in [14, Proposition 3.1.10], axiom ( ciw ) is derivable from mbC plus axiom ( ci ) . This being so, by Example 3.6 and the fact that ⊥ def = ( α ∧ ¬ α ) ∧¬ ( α ∧ ¬ α ) is a bottom formula in Cila (hence in
RCila ) for any α (i.e., ⊥ implies any otherformula), the connective ◦ could be eliminated from the signature of RCila , and so the logic
RCila could be defined over the signature Σ C (recall Definition 2.2). In that case, RCila would coincide with R C , the extension of C by adding the inference rule ( R ¬ ) (and wherethe notion of derivation is given as in Definition 2.6). The question is to find a model of RCila (or, equivalently, of R C ) which is still paraconsistent.In [21, Theorem 3.51(iv)] it was proved that ( IpE ) cannot hold in any paraconsistentextension of Ci in which the schema ( dm ) : ¬ ( α ∧ β ) → ( ¬ α ∨ ¬ β ) is valid. On the otherhand, in [14, Theorem 3.6.4] it was proved that the logic obtained from mbCciw by adding ( ca ∧ ) is equivalent to the logic obtained from mbCciw by adding the schema axiom ( dm ) .Since RCila derives ( ciw ) and ( ca ∧ ) , it also derives the schema ( dm ) . Given that RCila isan extension of Ci which satisfies ( IpE ) , it is not paraconsistent, by [21, Theorem 3.51(iv)]. Despite being very useful for finding models and countermodels, as it was shown in theprevious section, BALFI semantics does not seem to produce a decision procedure for
LFI swith replacement. In this section we will introduce a particular case of BALFIs based onpowerset Boolean algebras, which are more amenable to being generated by computationalmeans. These structures are in fact equivalent to neighborhood frames for non-normal modallogics, as we shall see in Section 6. Moreover, we shall prove in that Section that, with thissemantics,
RmbC can be defined within the bimodal version of the minimal modal logic E (a.k.a. classical modal logic , see [22, Definition 8.1]). Definition 5.1
Let W be a non-empty set. A neighborhood frame for RmbC over W is atriple F = h W, S ¬ , S ◦ i such that S ¬ : ℘ ( W ) → ℘ ( W ) and S ◦ : ℘ ( W ) → ℘ ( W ) are functions.A neighborhood model for RmbC over F is a pair M = hF , d i such that F is a neighborhoodframe for RmbC over W and d : V → ℘ ( W ) is a (valuation) function. Definition 5.2
Let M = hF , d i be a neighborhood model for RmbC over F = h W, S ¬ , S ◦ i .It induces a denotation function [[ · ]] M : F or (Σ) → ℘ ( W ) defined recursively as follows (bysimplicity, we will write [[ ϕ ]] instead of [[ ϕ ]] M when M is clear from the context): [[ p ]] = d ( p ) , if p ∈ V ; ϕ ∧ ψ ]] = [[ ϕ ]] ∩ [[ ψ ]] ; [[ ϕ ∨ ψ ]] = [[ ϕ ]] ∪ [[ ψ ]] ; [[ ϕ → ψ ]] = [[ ϕ ]] → [[ ψ ]] = ( W \ [[ ϕ ]]) ∪ [[ ψ ]] ; [[ ¬ ϕ ]] = ( W \ [[ ϕ ]]) ∪ S ¬ ([[ ϕ ]]) ; and [[ ◦ ϕ ]] = ( W \ ([[ ϕ ]] ∩ [[ ¬ ϕ ]])) ∩ S ◦ ([[ ϕ ]]) = ( W \ ([[ ϕ ]] ∩ S ¬ ([[ ϕ ]]))) ∩ S ◦ ([[ ϕ ]]) . Clearly [[ ϕ ]] ∪ [[ ¬ ϕ ]] = W , but [[ ϕ ]] ∩ [[ ¬ ϕ ]] is not necessarily empty. In addition, [[ ϕ ]] ∩ [[ ¬ ϕ ]] ∩ [[ ◦ ϕ ]] = ∅ . Definition 5.3
Let M = hF , d i be a neighborhood model for RmbC .(1) We say that a formula ϕ is valid (or true ) in M , denoted by M (cid:13) ϕ , if [[ ϕ ]] = W .(2) We say that a formula ϕ is valid w.r.t. neighborhood models , denoted by | = NM ϕ , if M (cid:13) ϕ for every neighborhood model M for RmbC .(3) The consequence relation | = NM induced by neighborhood models for RmbC is defined asfollows: Γ | = NM ϕ if either ϕ is valid w.r.t. neighborhood models for RmbC , or there existsa finite, non-empty subset { γ , . . . , γ n } of Γ such that ( γ ∧ ( γ ∧ ( . . . ∧ ( γ n − ∧ γ n ) . . . ))) → ϕ is valid w.r.t. neighborhood models for RmbC . Cleary, Γ | = NM ϕ if either ϕ is valid w.r.t. neighborhood models for RmbC , or thereexists a finite, non-empty subset { γ , . . . , γ n } of Γ such that, for every neighborhood model M for RmbC , T ni =1 [[ γ i ]] ⊆ [[ ϕ ]]. Proposition 5.4
Given a neighborhood frame F = h W, S ¬ , S ◦ i for RmbC let ˜ ¬ , ˜ ◦ : ℘ ( W ) → ℘ ( W ) defined as follows: ˜ ¬ ( X ) = ( W \ X ) ∪ S ¬ ( X ) and ˜ ◦ ( X ) = ( W \ ( X ∩ S ¬ ( X )) ∩ S ◦ ( X ) . Then B F def = h ℘ ( W ) , ∩ , ∪ , → , ˜ ¬ , ˜ ◦ , ∅ , W i is a BALFI. Moreover, if M = hF , d i is aneighborhood model for RmbC over F = h W, S ¬ , S ◦ i then the denotation function [[ · ]] M is avaluation over B F . Proof.
It is immediate from the definitions. ✷ Corollary 5.5 (Soundness of RmbC w.r.t. neighborhood models) If Γ ⊢ RmbC ϕ then Γ | = NM ϕ . Proof.
It follows from soundness of
RmbC w.r.t. BALFI semantics (Theorem 2.12) and byProposition 5.4. ✷ Proposition 5.4 suggests the following :
Definition 5.6
Let F = h W, S ¬ , S ◦ i be a neighborhood frame for RmbC . A formula ϕ is valid in F if M (cid:13) ϕ for every neighborhood model M = hF , d i for RmbC over F . In order to prove completeness of
RmbC w.r.t. neighborhood models, Stone’s represen-tation theorem for Boolean algebras will be used. This important theorem states that everyBoolean algebra is isomorphic to a Boolean subalgebra of ℘ ( W ), for a suitable W . Thismeans that, given a Boolean algebra A , there exists a set W and an injective homomorphism i : A → ℘ ( W ) of Boolean algebras. Note that i ( a ) = W if and only if a = 1.15 heorem 5.7 (Completeness of RmbC w.r.t. neighborhood models) If Γ | = NM ϕ then Γ ⊢ RmbC ϕ . Proof.
Let A can be the Boolean algebra with domain A can = F or (Σ) / ≡ , as defined in theproof of Theorem 2.13, and let B can be the corresponding expansion to Σ e . Let i : A can → ℘ ( W ) be an injective homomorphism of Boolean algebras, according to Stone’s theorem asdiscussed above. Consider the neighborhood frame F can = h W, S ¬ , S ◦ i for RmbC such thatthe functions S ¬ and S ◦ satisfy the following: S ¬ ( i ([ α ])) = i ([ ¬ α ]), and S ◦ ( i ([ α ])) = i ([ ◦ α ]),for every formula α (observe that these funcions are well-defined, since every connective in RmbC is congruential and i is injective). The values of these functions outside the imageof i are arbitrary. For instance, we can define S ¬ ( X ) = S ◦ ( X ) = ∅ if X / ∈ i [ A can ]. Now, let M can = hF can , d can i be the neighborhood model for RmbC such that d can ( p ) def = i ([ p ]), forevery propositional variable p . Fact: [[ α ]] = i ([ α ]), for every formula α .The proof of the Fact will be done by induction on the complexity of the formula α . Byconvenience, and as it is usually done (see, for instance, [14]), the complexity of ◦ α is definedto be stricty greater than the complexity of ¬ α . The case for α atomic or α = β γ for ∈ {∧ , ∨ , →} is clear, by the very definitions and by induction hypothesis. Now, supposethat α = ¬ β . By induction hypothesis, [[ β ]] = i ([ β ]). Observe that ∼ [ β ] ≤ [ ¬ β ] in A can (where ∼ denotes the Boolean complement in A can ), hence W \ i ([ β ]) = i ( ∼ [ β ]) ⊆ i ([ ¬ β ]).Thus,[[ ¬ β ]] = ( W \ [[ β ]]) ∪ S ¬ ([[ β ]]) = ( W \ i ([ β ])) ∪ S ¬ ( i ([ β ]))= ( W \ i ([ β ])) ∪ i ([ ¬ β ]) = i ([ ¬ β ]).Finally, let α = ◦ β . Since [ ◦ β ] ≤ ∼ ([ β ] ∧ [ ¬ β ]) in A can then i ([ ◦ β ]) ⊆ W \ ( i ([ β ]) ∩ i ([ ¬ β ])).Hence, by induction hypothesis,[[ ◦ β ]] = ( W \ ( i ([ β ]) ∩ i ([ ¬ β ]))) ∩ S ◦ ( i ([ β ])= ( W \ ( i ([ β ]) ∩ i ([ ¬ β ]))) ∩ i ([ ◦ β ]) = i ([ ◦ β ]).This concludes the proof of the Fact.Because of the Fact, M can (cid:13) α iff i ([ α ]) = W iff [ α ] = 1 iff ⊢ RmbC α . Now, suppose thatΓ | = NM ϕ . If | = NM ϕ then, in particular, M can (cid:13) ϕ and so ⊢ RmbC ϕ . From this, Γ ⊢ RmbC ϕ .On the other hand, suppose that there exists a finite, non-empty subset { γ , . . . , γ n } of Γsuch that | = NM ( γ ∧ ( γ ∧ ( . . . ∧ ( γ n − ∧ γ n ) . . . ))) → ϕ . By reasoning as above, it followsthat ⊢ RmbC ( γ ∧ ( γ ∧ ( . . . ∧ ( γ n − ∧ γ n ) . . . ))) → ϕ and so Γ ⊢ RmbC ϕ as well. ✷ RmbC is definable within the minimal bimodal modallogic
In this section it will be shown that
RmbC is definable within the bimodal version of theminimal modal logic E , also called classical modal logic in [22, Definition 8.1]). In terms ofcombination of modal logics, this bimodal logic is equivalent to the fusion (or, equivalently,the constrained fibring by sharing the classical connectives) of E with itself. This meansthat the minimal non-normal modal logic with two independent modalities (cid:3) and (cid:3) , whichwill be denoted by E ⊕ E , contains RmbC , the minimal self-extensional
LFI . As we shall see,both modalities are required for defining the two non-classical conectives ¬ and ◦ . Firstly,the definition of modal logic E will be briefly surveyed. Definition 6.1 ([22], Definition 7.1) A minimal model is a triple N = h W, N, d i suchthat W is a non-empty set and N : W → ℘ ( ℘ ( W )) and d : V → ℘ ( W ) are functions. Theclass of minimal models will we denoted by C M . Recall the signatures Σ m = {∧ , ∨ , → , ∼ , (cid:3) , ♦ } and Σ bm = {∧ , ∨ , → , ∼ , (cid:3) , ♦ , (cid:3) , ♦ } introduced in Definition 2.2. The class of models C M induces a modal consequence relationdefined as follows: Definition 6.2 ([22], Definition 7.2)
Let N be a minimal model and w ∈ W . N is saidto satisfy a formula ϕ ∈ F or (Σ m ) in w , denoted by | = N w ϕ , according to the following recursivedefinition (here [[ ϕ ]] N denotes the set { w ∈ W : | = N w ϕ } , the denotation of ϕ in N ):1. if p is a propositional variable then | = N w p iff w ∈ d ( p ) ;2. | = N w ∼ α iff = N w α ;3. | = N w α ∧ β iff | = N w α and | = N w β ;4. | = N w α ∨ β iff | = N w α or | = N w β ;5. | = N w α → β iff = N w α or | = N w β ;6. | = N w (cid:3) α iff [[ α ]] N ∈ N ( w ) ;7. | = N w ♦ α iff ( W \ [[ α ]] N ) / ∈ N ( w ) . A formula ϕ is true in N if [[ ϕ ]] N = W , and it is valid w.r.t. C M , denoted by | = C M ϕ ,if it is true in every minimal model. The degree-preserving consequence w.r.t. C M can bedefined analogously to the one for neighborhood semantics for RmbC given in Definition 5.3.Namely, Γ | = C M ϕ if either | = C M ϕ , or there exists a finite, non-empty subset { γ , . . . , γ n } ofΓ such that | = C M ( γ ∧ ( γ ∧ ( . . . ∧ ( γ n − ∧ γ n ) . . . ))) → ϕ . The latter is equivalent to say that T ni =1 [[ γ i ]] N ⊆ [[ ϕ ]] N . For the basic notions of combining logics the reader can consult [12, 16]. efinition 6.3 ([22], Definition 8.1) The minimal modal logic (or classical modal logic ) E is defined by means of a Hilbert calculus over the signature Σ m obtained by adding to theHilbert calculus for CPL + (recall Definition 2.3) the following axiom schemas and rules: α ∨ ∼ α ( PEM ) α → (cid:0) ∼ α → β (cid:1) ( exp ) ♦ α ↔ ∼ (cid:3) ∼ α ( AxMod ) α ↔ β (cid:3) α ↔ (cid:3) β ( R (cid:3) )The notion of derivations in E is defined as for RmbC , recall Definition 2.6. Note that(
PEM ) and ( exp ), together with
CPL + , guarantee that E is an expansion of propositionalclassical logic by adding the modalities (cid:3) and ♦ which are interdefinable as usual, and suchthat both are congruential. That is, E satisfies replacement. Theorem 6.4 ([22], Section 9.2)
The logic E is sound and complete w.r.t. the semanticsin C M , namely: Γ ⊢ E ϕ iff Γ | = C M ϕ . Definition 6.5 (Minimal bimodal logic)
The minimal bimodal logic E ⊕ E is defined bymeans of a Hilbert calculus over signature Σ bm obtained by adding to the Hilbert calculus for CPL + the following axiom schemas and rules, for i = 1 , : α ∨ ∼ α ( PEM ) α → (cid:0) ∼ α → β (cid:1) ( exp ) ♦ i α ↔ ∼ (cid:3) i ∼ α ( AxMod i ) α ↔ β (cid:3) i α ↔ (cid:3) i β ( R (cid:3) i )Observe that E ⊕ E is obtained from E by ‘duplicating’ the modalities. There is norelationship between (cid:3) and (cid:3) and so ♦ and ♦ are also independent.The semantics of E ⊕ E is given by the class C ′M of structures of the form N = h W, N , N , d i such that W is a non-empty set and N i : W → ℘ ( ℘ ( W )) (for i = 1 ,
2) and d : V → ℘ ( W ) arefunctions. The denotation [[ ϕ ]] N of a formula ϕ ∈ F or (Σ bm ) in N is defined by an obviousadaptation of Definition 6.2 to F or (Σ bm ). By defining the consequence relations ⊢ E ⊕ E and | = C ′M in analogy to the ones for E , it is straightforward to adapt the proof of soundness andcompleteness of E to the bimodal case: Theorem 6.6
The logic E ⊕ E is sound and complete w.r.t. the semantics in C ′M , namely: Γ ⊢ E ⊕ E ϕ iff Γ | = C ′M ϕ . E ⊕ E is the fusion (or, equivalently, theconstrained fibring by sharing the classical connectives) of E with itself. Finally, it will be shown that
RmbC can be defined inside E ⊕ E by means of the followingabbreviations: ¬ ϕ def = ϕ → (cid:3) ϕ and ◦ ϕ def = ∼ ( ϕ ∧ (cid:3) ϕ ) ∧ (cid:3) ϕ. In order to see this, observe that any function N : W → ℘ ( ℘ ( W )) induces a unique function S : ℘ ( W ) → ℘ ( W ) given by S ( X ) = { w ∈ W : X ∈ N ( w ) } . Conversely, any function S : ℘ ( W ) → ℘ ( W ) induces a function N : W → ℘ ( ℘ ( W )) given by N ( w ) = { X ⊆ W : w ∈ S ( X ) } . Both functions are inverses of each other. From this, a structure (or minimalmodel) N = h W, N , N , d i for E ⊕ E can be transformed into a neighborhood model M = h W, S ¬ , S ◦ , d i for RmbC such that S ¬ and S ◦ are obtained, respectively, from the functions N and N as indicated above. Observe that w ∈ [[ (cid:3) ϕ ]] N iff | = N w (cid:3) ϕ iff [[ ϕ ]] N ∈ N ( w ) iff w ∈ S ¬ ([[ ϕ ]] N ) . That is, S ¬ ([[ ϕ ]] N ) = [[ (cid:3) ϕ ]] N . Analogously, S ◦ ([[ ϕ ]] N ) = [[ (cid:3) ϕ ]] N . From this, it is easy toprove by induction on the complexity of the formula ϕ ∈ F or (Σ) that [[ ϕ ]] M = [[ ϕ t ]] N , where ϕ t is the formula over the signature Σ bm obtained from ϕ by replacing any ocurrence of theconnectives ¬ and ◦ by the corresponding abbreviations, as indicated above. Conversely,any neighborhood model M = h W, S ¬ , S ◦ , d i for RmbC gives origin to a unique minimalmodel N = h W, N , N , d i for E ⊕ E such that [[ ϕ ]] M = [[ ϕ t ]] N for every formula ϕ ∈ F or (Σ).That is, the class of minimal models for E ⊕ E coincides (up to presentation) with the classof neighborhood models for RmbC , and both classes validate the same formulas over thesignature Σ of
RmbC . From this, Corollary 5.5, Theorem 5.7 and Theorem 6.6 we show that
RmbC is definable within E ⊕ E : Theorem 6.7
The logic
RmbC is definable within E ⊕ E , in the following sense: Γ ⊢ RmbC ϕ iff Γ t ⊢ E ⊕ E ϕ t for every Γ ∪ { ϕ } ⊆ F or (Σ) , where Γ t = { ψ t : ψ ∈ Γ } . The main result obtained in this section, namely Theorem 6.7, establishes an interestingrelation between non-normal modal logics and paraconsistent logics. Connections betweenmodalities and paraconsistency are well-known in the literature. In [7, 8], for instance,B´eziau proposes to consider a paraconsistent negation defined in the modal system S5 as ¬ ϕ def = ♦ ∼ ϕ . This way of defining a paraconsistent negation inside a modal logic has beenalready regarded in 1987 in [27], when a Kripke-style semantics was proposed for Sette’s three-valued paraconsistent logic P1 based on Kripke frames for the modal logic T . This resultwas improved in [20], by showing that P1 can be interpreted in T by means of Kripke frameshaving at most two worlds. Moreover, in 1982 Segerberg already suggested in [38, p. 128]the possibility of studying the (unexplored at that time) modal notion of ‘ ϕ is non-necesary’,namely ∼ (cid:3) ϕ (which is of course equivalent in most modal systems to ♦ ∼ ϕ ). Several authorshave explored the possibility of defining such paraconsistent negation in other modal logicssuch as B [4], S4 [25] and even weaker modal systems [11]. In such context, Marcos proposesin [33], besides the paraconsistent negation defined as above, the definition of a consistencyconnective within a modal system by means of the formula ◦ ϕ def = ϕ → (cid:3) ϕ (observe the See [12, 16]. E ⊕ E ). In that paperit is shown that any normal modal logic in which the schema ϕ → (cid:3) ϕ is not valid givesorigin to an LFI in this way. Moreover, it is shown that it is also possible to start from a“modal
LFI ”, over the signature Σ of
LFI s, in which the paraconsistent negation and theconsistency connective enjoy a Kripke-style semantics, defining the modal necessity operatorby means of the formula (cid:3) ϕ def = ∼¬ ϕ (where ∼ is the strong negation defined as in mbC ,recall Example 3.6). This shows that ‘reasonable’ normal modal logics and LFI s are twofaces of the same coin. Our Theorem 6.7 partially extends this relationship to the realmof non-normal modal logics. The result we have obtained is partial, in the sense that theminimum bimodal non-normal modal logics gives origin to
RmbC , but the converse doesnot seem to be true. Namely, starting from
RmbC it is not obvious that the modalities (cid:3) and (cid:3) could be defined by means of formulas in the signature Σ. This topic deserves furtherinvestigation. Recall the axioms considered in Definition 3.1. Because of the limit to paraconsistencyimposed by
RCila (recall Example 4.2), in this section Ax will denote a set formed by oneor more of the axiom schemas introduced in Definition 3.1 with the exception of ( ca ) for ∈ {∧ , ∨ , →} . Let NM ( Ax ) the class of neighborhood frames in which every schema in Ax is valid. Define the consequence relation | = NM ( Ax ) in the obvious way. By adapting theprevious results it is easy to prove the following: Theorem 7.1 (Soundness and completeness of RmbC ( Ax ) w.r.t. NM ( Ax ) ) Let Γ ∪ { ϕ } ⊆ F or (Σ) . Then: Γ ⊢ RmbC ( Ax ) ϕ if and only if Γ | = NM ( Ax ) ϕ . The class of neighborhood frames which validates each of the axioms of Ax can be easilycharacterized: Proposition 7.2
Let F be a neighborhood frame for RmbC .Then:(1) ( ciw ) is valid in F iff W \ ( X ∩ S ¬ ( X )) ⊆ S ◦ ( X ) , for every X ⊆ W ;(2) ( ci ) is valid in F iff W \ ( X ∩ S ¬ ( X )) ⊆ S ◦ ( X ) \ S ¬ (( W \ ( X ∩ S ¬ ( X ))) ∩ S ◦ ( X )) , forevery X ⊆ W ;(3) ( cl ) is valid in F iff S ¬ ( X ∩ S ¬ ( X )) ⊆ W \ ( X ∩ S ¬ ( X )) ⊆ S ◦ ( X ) , for every X ⊆ W ;(4) ( cf ) is valid in F iff ( X \ S ¬ ( X )) ∪ S ¬ ( X \ S ¬ ( X )) ⊆ X , for every X ⊆ W ;(5) ( ce ) is valid in F iff X ⊆ ( X \ S ¬ ( X )) ∪ S ¬ ( X \ S ¬ ( X )) , for every X ⊆ W . Recall the minimal bimodal logic E ⊕ E studied in Section 6. If ϕ is a formula in F or (Σ bm )then E ⊕ E ( ϕ ) will denote the extension of E ⊕ E by adding ϕ as an axiom schema. Let C ′M ( ϕ ) be the class of structures (i.e., minimal models) N for E ⊕ E such that ϕ is valid in N (as an axiom schema). Theorem 6.6 can be extended to prove that the logic E ⊕ E ( ϕ ) issound and complete w.r.t. the semantics in C ′M ( ϕ ). From this, and taking into account the20epresentability of RmbC within E ⊕ E (Theorem 6.7) and the equivalence between minimalmodels for E ⊕ E and neighborhood models for RmbC discussed right before Theorem 6.7,Proposition 7.2 can be recast as follows:
Corollary 7.3 (1)
RmbC ( ciw ) is definable in E ⊕ E ( ∼ ( ϕ ∧ (cid:3) ϕ ) → (cid:3) ϕ ) ;(2) RmbC ( ci ) is definable in E ⊕ E ( ∼ ( ϕ ∧ (cid:3) ϕ ) → ( (cid:3) ϕ ∧ ∼ (cid:3) ( ∼ ( ϕ ∧ (cid:3) ϕ ) ∧ (cid:3) ϕ ))) or,equivalenty, in E ⊕ E (( (cid:3) ϕ → (cid:3) ( ∼ ( ϕ ∧ (cid:3) ϕ ) ∧ (cid:3) ϕ )) → ( ϕ ∧ (cid:3) ϕ )) ;(3) RmbC ( cl ) is definable in E ⊕ E (( (cid:3) ( ϕ ∧ (cid:3) ϕ ) → ∼ ( ϕ ∧ (cid:3) ϕ )) ∧ ( ∼ ( ϕ ∧ (cid:3) ϕ ) → (cid:3) ϕ )) ;(4) RmbC ( cf ) is definable in E ⊕ E ((( ϕ ∧ ∼ (cid:3) ϕ ) ∨ (cid:3) ( ϕ ∧ ∼ (cid:3) ϕ )) → ϕ ) ;(5) RmbC ( ce ) is definable in E ⊕ E ( ϕ → (( ϕ ∧ ∼ (cid:3) ϕ ) ∨ (cid:3) ( ϕ ∧ ∼ (cid:3) ϕ ))) . As it was mentioned in Section 2, the BALFI semantics for
RmbC , as well as its neigh-borhood semantics presented in Section 5, is degree-preserving instead of truth-preserving(using the terminology from [10]). This requires adapting, in a coherent way, the usual defi-nition of derivation from premises in a Hilbert calculus, recall Definition 2.6. This is exactlythe methodology adopted with most normal modal logics in which the semantics is local,thus recovering the deduction metatheorem. But it is also possible to consider global (ortruth-preserving) semantics, as it is usually done with algebraic semantics. This leads us toconsider the logic
RmbC ∗ , which is defined by the same Hilbert calculus than the one for RmbC , but now derivations from premises in
RmbC ∗ are defined as usual in Hilbert calculi. Definition 8.1
The logic
RmbC ∗ is defined by the same Hilbert calculus over signature Σ than RmbC , that is, by adding to mbC the inference rules ( R ¬ ) and ( R ◦ ). Definition 8.2 (Derivations in RmbC ∗ ) We say that a formula ϕ is derivable in RmbC ∗ from Γ , and we write Γ ⊢ RmbC ∗ ϕ , if there exists a finite sequence of formulas ϕ . . . ϕ n suchthat ϕ n is ϕ and, for every ≤ i ≤ n , either ϕ i is an instance of an axiom of RmbC , or ϕ i ∈ Γ , or ϕ i is the consequence of some inference rule of RmbC whose premises appear inthe sequence ϕ . . . ϕ i − . Now, the degree-preserving BALFI semantics for
RmbC given in Definition 2.10 mustbe replaced by a truth-preserving consequence relation for
RmbC ∗ : Definition 8.3 (Truth-preserving BALFI semantics)
Let Γ ∪ { ϕ } be a set of formulas in F or (Θ) . We say that ϕ is a global (or truth-preserving )consequence of Γ in BI , denoted by Γ | = g BI ϕ , if either ϕ is valid in BI , or there exists a finite,non-empty subset { γ , . . . , γ n } of Γ such that, for every BALFI B and every valuation v overit, if v ( γ i ) = 1 for every ≤ i ≤ n then v ( ϕ ) = 1 . The proof of the following result follows by an easy adaptation of the proof of soundnessand completeness of
RmbC w.r.t. BALFI semantics:21 heorem 8.4 (Soundness and completeness of RmbC ∗ w.r.t. truth-preservingsemantics) For every Γ ∪ { ϕ } ⊆ F or (Θ) : Γ ⊢ RmbC ∗ ϕ iff Γ | = g BI ϕ . Remark 8.5
The definition of truth-preserving semantics restricts the number of paracon-sistent models for
RmbC ∗ . Indeed, let p and q be two different propositional variables. Inorder to show that p, ¬ p = g BI q , there must be a BALFI B and a valuation v over B suchthat v ( p ) = v ( ¬ p ) = 1 but v ( q ) = 1 . That is, B must be such that ¬ . Since ¬ , itfollows that ¬¬ ¬ in B . This shows that there is no paraconsistent extensionof RmbC ∗ which satisfies axiom ( cf ) . In particular, there is no paraconsistent extension of RmbC ∗ satisfying axioms ( cf ) and ( ci ) . Thus, the open problems solved in Examples 3.8and 3.9 have a negative answer in this setting. This shows that the truth-preserving approachis much more restricted than the degree-preserving approach in terms of paraconsistency.In any case, there are still paraconsistent BALFIs for the truth-preserving logic RmbC ∗ (namely, the ones such that ¬ ). The situation is quite different in the realm of fuzzylogics: in [23, 28], among others, it was studied the degree-preserving companion of sev-eral fuzzy logics, showing that their usual truth-preserving consequence relations are neverparaconsistent. The distinction between local and global reasoning has been studied by A. Sernadas andhis collaborators (for a brief exposition see, for instance, [16], Section 2.3 in Chapter 2).From the proof-theoretical perspective, the Hilbert calculi (called
Hilbert calculi with carefulreasoning in [16, Definition 2.3.1]) are of the form H = h Θ , R g , R l i where Θ is a propositionalsignature and R g ∪ R l is a set of inference rules such that R l ⊆ R g and no element of R g \ R l is an axiom schema. Elements of R g and R l are called global and local inference rules,respectively. Given Γ ∪ { ϕ } ⊆ F or (Θ), ϕ is globally derivable from Γ in H , written Γ ⊢ gH ϕ ,if ϕ is derivable from Γ in the Hilbert calculus h Θ , R g i by using the standard definition (seeDefinition 8.2). On the other hand, in local derivations, besides using the local rules andthe premises, global rules can be used provided that the premises are (global) theorems. Informal terms, ϕ is locally derivable from Γ in H , written Γ ⊢ lH ϕ , if there exists a finitesequence of formulas ϕ . . . ϕ n such that ϕ n is ϕ and, for every 1 ≤ i ≤ n , either ϕ i ∈ Γ, or ⊢ gH ϕ i , or ϕ i is the consequence of some inference rule of R l whose premises appear in thesequence ϕ . . . ϕ i − (observe that this includes the case when ϕ i is an instance of an axiomin R l ). Obviously, local derivations are global derivations, and local and global theoremscoincide.For instance, typically a Hilbert calculus for a (normal) modal logic contains, as localinference rules, ( MP ) and the axiom schemas, while the set of global rules is ( R l ) plusthe Necessitation rule. As we have seen in Section 6, the same is the case for minimalnon-normal modal logics, but with Replacement for (cid:3) instead of Necessitation. In thiscase, the deduction metatheorem only holds for local derivations. Note that, by definition,derivations in RmbC ∗ lie in the scope of global derivations, while derivations in RmbC are local derivations. Hence, the extension of mbC with replacement can be recast as aHilbert calculus with careful reasoning
RmbC + = h Σ , R g , R l i such that R l contains theaxiom schemas of mbC plus ( MP ), and R g contains, besides this, the rules ( R ¬ ) and ( R ◦ ).Of course the same can be done with the axiomatic extensions of mbC (and so of RmbC )considered in Section 3. 22t the semantical level, local derivations correspond to degree-preserving semantics w.r.t.a given class M of algebras, while global derivations correspond to truth-preserving semanticsw.r.t. the class M .The presentation of LFI s with replacement as Hilbert calculi with careful reasoning (asthe case of
RmbC + ) can be useful in order to combine these logics with (standard) normalmodal logics by algebraic fibring: in this case, completeness of the fibring of the correspondingHilbert calculi w.r.t. a semantics given by classes of suitable expansions of Boolean algebraswould be immediate, according to the results stated in [16, Chapter 2]. By considering, asdone in [41], classes M of powerset algebras (i.e., with domain of the form ℘ ( W ) for a non-empty set W ) induced by Kripke models (which can be generalized to neighborhood models),then the fibring of, say, RmbC + with a given modal logic would simply be a minimal logic E with three primitive modalities ( (cid:3) , (cid:3) , and (cid:3) ), from which we derive the followingmodalities: ♦ ϕ def = ∼ (cid:3) ∼ ϕ , ¬ ϕ def = ϕ → (cid:3) ϕ , and ◦ ϕ def = ∼ ( ϕ ∧ (cid:3) ϕ ) ∧ (cid:3) ϕ . This opensinteresting opportunities for future research. The next step is extending
RmbC , as well as its axiomatic extensions analyzed above, tofirst-order languages. In order to do this, we will adapt our previous approach to quantified
LFI s, see [18], [14, Chapter 7], [24]) to this framework. To begin with, the first-order version
RQmbC of RmbC will be introduced.
Definition 9.1
Let
V ar = { v , v , . . . } be a denumerable set of individual variables. A first-order signature Ω is given as follows:- a set C of individual constants;- for each n ≥ , a set F n of function symbols of arity n ,- for each n ≥ , a nonempty set P n of predicate symbols of arity n . The sets of terms and formulas generated by a signature Ω (with underlying propositionalsignature Σ) will be denoted by
T er (Ω) and
F or (Ω), respectively. The set of closed formulas(or sentences) and the set of closed terms (terms without variables) over Ω will be denotedby Sen (Ω) and
CT er (Ω), respectively. The formula obtained from a given formula ϕ bysubstituting every free occurrence of a variable x by a term t will be denoted by ϕ [ x/t ]. Definition 9.2
Let Ω be a first-order signature. The logic RQmbC is obtained from
RmbC by adding the following axioms and rules:
Axiom Schemas: Ax ∃ ) ϕ [ x/t ] → ∃ xϕ, if t is a term free for x in ϕ ( Ax ∀ ) ∀ xϕ → ϕ [ x/t ] , if t is a term free for x in ϕ Inference rules: ( ∃ -In ) ϕ → ψ ∃ xϕ → ψ , where x does not occur free in ψ ( ∀ -In ) ϕ → ψϕ → ∀ xψ , where x does not occur free in ϕ The consequence relation of
RQmbC , adapted from the one for
RmbC (recall Defini-tion 2.6) will be denoted by ⊢ RQmbC . Remarks 9.3 (1) It is worth mentioning that the only difference between
QmbC and
RQmbC is that thelatter contains the inference rules ( R ¬ ) and ( R ◦ ) , which are not present in the former (besidesthe different notions of derivation from premisses adopted in QmbC and in
RQmbC ).(2) Recall that a Hilbert calculus with careful reasoning for
RmbC called
RmbC + was definedat the end of Section 8. This can extended to RQmbC by considering the Hilbert calculuswith careful reasoning
RQmbC + over a given first-order signature Ω , such that R l containsthe axiom schemas of QmbC (over Ω ) plus ( MP ) , and R g contains, besides this, the rules ( R ¬ ) , ( R ◦ ) , ( ∃ -In ) and ( ∀ -In ) (over Ω ).
10 BALFI semantics for RQmbC
In [24] a semantics of first-order structures based on swap structures over complete Booleanalgebras was obtained for
QmbC , a first-order version of mbC proposed in [18]. Since
RQmbC is self-extensional, that semantics can be drastically simplified, and so the non-de-terministic swap structures will be replaced by BALFIs, which are ordinary algebras. Fromnow on, only BALFIs over complete Boolean algebras will be considered.
Definition 10.1 A complete BALFI is a BALFI such that its reduct to Σ BA is a completeBoolean algebra. Definition 10.2
Let B be a complete BALFI, and let Ω be a first-order signature. A (first-order) structure over B and Ω (or a RQmbC -structure over Ω ) is a pair A = h U, I A i suchthat U is a nonempty set (the domain or universe of the structure) and I A is an interpretationfunction which assigns:- an element I A ( c ) of U to each individual constant c ∈ C ;- a function I A ( f ) : U n → U to each function symbol f of arity n ;- a function I A ( P ) : U n → A to each predicate symbol P of arity n . otation 10.3 From now on, we will write c A , f A and P A instead of I A ( c ) , I A ( f ) and I A ( P ) to denote the interpretation of an individual constant symbol c , a function symbol f and apredicate symbol P , respectively. Definition 10.4
Given a structure A over B and Ω , an assignment over A is any function µ : V ar → U . Definition 10.5
Given a structure A over B and Ω , and given an assignment µ : V ar → U we define recursively, for each term t , an element [[ t ]] A µ in U as follows:- [[ c ]] A µ = c A if c is an individual constant;- [[ x ]] A µ = µ ( x ) if x is a variable;- [[ f ( t , . . . , t n )]] A µ = f A ([[ t ]] A µ , . . . , [[ t n ]] A µ ) if f is a function symbol of arity n and t , . . . , t n are terms. Definition 10.6
Let A be a structure over B and Ω . The diagram language of A is the set offormulas F or (Ω U ) , where Ω U is the signature obtained from Ω by adding, for each element u ∈ U , a new individual constant ¯ u . Definition 10.7
The structure b A = h U, I b A i over Ω U is the structure A over Ω extended by I b A (¯ u ) = u for every u ∈ U . It is worth noting that s b A = s A whenever s is a symbol (individual constant, function symbolor predicate symbol) of Ω. Notation 10.8
The set of sentences or closed formulas (that is, formulas without free vari-ables) of the diagram language
F or (Ω U ) is denoted by Sen (Ω U ) , and the set of terms and ofclosed terms over Ω U will be denoted by T er (Ω U ) and CT er (Ω U ) , respectively. If t is a closedterm we can write [[ t ]] A instead of [[ t ]] A µ , for any assignment µ , since it does not depend on µ . Definition 10.9 (RQmbC interpretation maps)
Let B be a complete BALFI, and let A be a structure over B and Ω . The interpretation map for RQmbC over A and B is afunction [[ · ]] A : Sen (Ω U ) → A satisfying the following clauses:( i ) [[ P ( t , . . . , t n )]] A = P A ([[ t ]] b A , . . . , [[ t n ]] b A ) , if P ( t , . . . , t n ) is atomic;( ii ) [[ ϕ ]] A = ϕ ]] A , for every ∈ {¬ , ◦} ;( iii ) [[ ϕ ψ ]] A = [[ ϕ ]] A ψ ]] A , for every ∈ {∧ , ∨ , →} ;( iv ) [[ ∀ xϕ ]] A = V u ∈ U [[ ϕ [ x/ ¯ u ]]] A ;( v ) [[ ∃ xϕ ]] A = W u ∈ U [[ ϕ [ x/ ¯ u ]]] A . Recall the notation stated in Definition 10.6. The interpretation map can be extended toarbitrary formulas as follows: 25 efinition 10.10
Let B be a complete BALFI, and let A be a structure over B and Ω .Given an assignment µ over A , the extended interpretation map [[ · ]] A µ : F or (Ω U ) → A isgiven by [[ ϕ ]] A µ = [[ ϕ [ x /µ ( x ) , . . . , x n /µ ( x n )]]] A , provided that the free variables of ϕ occur in { x , . . . , x n } . For every u ∈ U and every assignment µ , let µ xu be the assignment such that µ xu ( x ) = u and µ xu ( y ) = µ ( y ) if y = x . Then, it is immediate to see that [[ ϕ ]] A µ xu = [[ ϕ [ x/ ¯ u ]]] A µ , for everyformula ϕ . Definition 10.11
Let B be a complete BALFI, and let A be a structure over B and Ω .(1) Given a formula ϕ in F or (Ω U ) , ϕ is said to be valid in ( A , B ) , denoted by | = ( A , B ) ϕ , if [[ ϕ ]] A µ = 1 , for every assignment µ .(2) Given a set of formulas Γ ∪ { ϕ } ⊆ F or (Ω U ) , ϕ is said to be a semantical consequence ofΓ w.r.t. ( A , B ) , denoted by Γ | = ( A , B ) ϕ , if either ϕ is valid in ( A , B ) , or there exists a finite,non-empty subset { γ , . . . , γ n } of Γ such that the formula ( γ ∧ ( γ ∧ ( . . . ∧ ( γ n − ∧ γ n ) . . . ))) → ϕ is valid in ( A , B ) . Definition 10.12 (First-order degree-preserving BALFI semantics)
Let Γ ∪ { ϕ } ⊆ F or (Ω) be a set of formulas. Then ϕ is said to be a semantical consequence of Γ in RQmbC w.r.t. BALFIs , denoted by Γ | = RQmbC ϕ , if Γ | = ( A , B ) ϕ for every pair ( A , B ) . As in the case of
RmbC , given that
RQmbC uses local reasoning, it satisfies the deduc-tion metatheorem without any restrictions. This is different to what happens with
QmbC ,where this metatheorem holds with the same restrictions than in first-order classical logic.
Theorem 10.13 (Deduction Metatheorem for RQmbC) Γ , ϕ ⊢ RQmbC ψ if and only if Γ ⊢ RQmbC ϕ → ψ . In order to prove the soundness of
RQmbC w.r.t. BALFI semantics, it is necessary tostate an important result:
Theorem 10.14 (Substitution Lemma)
Let B be a complete BALFI, A a structure over B and Ω , and µ an assignment over A . If t is a term free for z in ϕ and b = [[ t ]] b A µ , then [[ ϕ [ z/t ]]] A µ = [[ ϕ [ z/ ¯ b ]]] A µ . Proof.
It is proved by induction on the complexity of ϕ . ✷ Theorem 10.15 (Soundness of RQmbC w.r.t. BALFIs)
For every set Γ ∪ { ϕ } ⊆ F or (Ω) : Γ ⊢ RQmbC ϕ implies that Γ | = RQmbC ϕ . Proof.
It will be proven by extending the proof of soundness of
RmbC w.r.t. BALFIsemantics (Theorem 2.12). Thus, the only cases required to be analyzed are the new axiomsand inference rules. By the very definitions, and taking into account Theorem 10.14, itis immediate to see that axioms ( Ax ∃ ) and ( Ax ∀ ) are valid in any ( A , B ). With respectto ( ∃ -In ), suppose that α → β is valid in a given ( A , B ), where the variable x does notoccur free in β . Then [[ α ]] A µ ≤ [[ β ]] A µ for every assignment µ . In particular, for every u ∈ U ,[[ α ]] A µ xu ≤ [[ β ]] A µ xu = [[ β ]] A µ , since x is not free in β . But then: [[ ∃ xα ]] A µ = W u ∈ U [[ α [ x/ ¯ u ]]] A µ = W u ∈ U [[ α ]] A µ xu ≤ [[ β ]] A µ . Hence, ∃ xα → β is valid in ( A , B ). The case for ( ∀ -In ) is provedanalogously. ✷ This section is devoted to prove the completeness of
RQmbC w.r.t. BALFI semantics. Theproof will be an adaptation to the completeness proof for
QmbC w.r.t. swap structuressemantics given in [24].
Definition 11.1
Consider a theory ∆ ⊆ F or (Ω) and a nonempty set C of constants of thesignature Ω . Then, ∆ is called a C - Henkin theory in RQmbC if it satisfies the following:for every formula ϕ with (at most) a free variable x , there exists a constant c in C such that ∆ ⊢ RQmbC ∃ xϕ → ϕ [ x/c ] . Remark 11.2
As observed in [24], it is easy to show that, if ∆ is a C -Henkin theory in QmbC and ϕ is a formula with (at most) a free variable x then there is a constant c in C such that ∆ ⊢ QmbC ϕ [ x/c ] → ∀ xϕ . Of course the same result holds for RQmbC . Definition 11.3
Let Ω C be the signature obtained from Ω by adding a set C of new individualconstants. The consequence relation ⊢ C RQmbC is the consequence relation of
RQmbC overthe signature Ω C . Recall that, given a Tarskian and finitary logic L = h F or, ⊢i (where F or is the set offormulas of L ), and given a set Γ ∪ { ϕ } ⊆ F or , the set Γ is said to be maximally non-trivialwith respect to ϕ in L if the following holds: (i) Γ ϕ , and (ii) Γ , ψ ⊢ ϕ for every ψ / ∈ Γ.By straightforwardly adapting [24, Proposition 8.4] from
QmbC to RQmbC , we obtain thefollowing:
Proposition 11.4
Let Γ ∪ { ϕ } ⊆ Sen (Ω) such that Γ RQmbC ϕ . Then, there exists a setof formulas ∆ ⊆ F or (Ω C ) , for some nonempty set C of new individual constants, such that Γ ⊆ ∆ , ∆ is a C -Henkin theory in RQmbC and, in addition, ∆ is maximally non-trivialwith respect to ϕ in RQmbC . Definition 11.5
Consider a set ∆ ⊆ F or (Ω) which is non-trivial in RQmbC , that is:there is some formula ϕ in F or (Ω) such that ∆ RQmbC ϕ . Let ≡ ∆ ⊆ F or (Ω) be therelation in F or (Ω) defined as follows: α ≡ ∆ β iff ∆ ⊢ RQmbC α ↔ β . By adapting the proof of Theorem 2.13 it follows that ≡ ∆ is an equivalence relationwhich induces a Boolean algebra A ∆ def = h A ∆ , ∧ , ∨ , → , ∆ , ∆ i , where A ∆ def = F or (Ω) / ≡ ∆ ,[ α ] ∆ β ] ∆ def = [ α β ] ∆ for any ∈ {∧ , ∨ , →} , 0 ∆ def = [ ϕ ∧ ( ¬ ϕ ∧◦ ϕ )] ∆ and 1 ∆ def = [ ϕ ∨¬ ϕ ] ∆ .Moreover, by defining α ] ∆ def = [ α ] ∆ for any ∈ {¬ , ◦} we obtain a BALFI denoted by B ∆ .The construction of the canonical model for RQmbC w.r.t. ∆ requires a completeBALFI, hence the Boolean algebra A ∆ must be completed. Recall that a Boolean algebra A ′ is a completion of a Boolean algebra A if: (1) A ′ is complete, and (2) A ′ includes A asa dense subalgebra (that is: every element in A ′ is the supremum, in A ′ , of some subset of A ). From this, A ′ preserves all the existing infima and suprema in A . In formal terms: there See, for instance, [29, Chapter 25]. ∗ : A → A ′ such that ∗ ( W A X ) = W A ′ ∗ [ X ] for every X ⊆ A such that the supremum W A X exists, where ∗ [ X ] = {∗ ( a ) : a ∈ X } . Analogously, ∗ ( V A X ) = V A ′ ∗ [ X ] for every X ⊆ A such thatthe infimum V A X exists. By the well-known results obtained independently by MacNeilleand Tarski, it follows that every Boolean algebra has a completion; moreover, the completionis unique up to isomorphisms. Based on this, let C A ∆ be the completion of A ∆ and let ∗ : A ∆ → C A ∆ be the associated monomorphism. Definition 11.6
Let C A ∆ be the complete Boolean algebra defined as above. The canonicalBALFI for RQmbC over ∆ , denoted by B ∆ , is obtained from C A ∆ by adding the unaryoperators ¬ and ◦ defined as follows: ¬ b = ∗ ( ¬ a ) if b = ∗ ( a ) , and ¬ b = ∼ b if b / ∈ ∗ [ A ∆ ] ; ◦ b = ∗ ( ◦ a ) if b = ∗ ( a ) , and ◦ b = 1 if b / ∈ ∗ [ A ∆ ] . Proposition 11.7
The operations over B ∆ are well-defined, and B ∆ is a complete BALFIsuch that ∗ ([ α ] ∆ ) = 1 iff ∆ ⊢ RQmbC α . Proof.
Since ∗ [ A ∆ ] is a subalgebra of C A ∆ , b / ∈ ∗ [ A ∆ ] iff ∼ b / ∈ ∗ [ A ∆ ]. On the other hand, ∗ is injective. This shows that ¬ and ◦ are well-defined. The rest of the proof is obvious fromthe definitions. ✷ Definition 11.8 (Canonical Structure) Let Ω be a signature with some individual constant.Let ∆ ⊆ F or (Ω) be non-trivial in RQmbC , let B ∆ be as in Definition 11.6, and let U = CT er (Ω) . The canonical structure induced by ∆ is the structure A ∆ = h U, I A ∆ i over B ∆ and Ω such that:- c A ∆ = c , for each individual constant c ;- f A ∆ : U n → U is such that f A ∆ ( t , . . . , t n ) = f ( t , . . . , t n ) , for each function symbol f of arity n ;- P A ∆ ( t , . . . , t n ) = ∗ ([ P ( t , . . . , t n )] ∆ ) , for each predicate symbol P of arity n . Definition 11.9
Let ( · ) ⊲ : ( T er (Ω U ) ∪ F or (Ω U )) → ( T er (Ω) ∪ F or (Ω)) be the mapping suchthat ( s ) ⊲ is the expression obtained from s by substituting every occurrence of a constant ¯ t by the term t itself, for t ∈ CT er (Ω) . Lemma 11.10
Let ∆ ⊆ F or (Ω) be a set of formulas over a signature Ω such that ∆ is a C -Henkin theory in RQmbC for a nonempty set C of individual constants of Ω , and ∆ ismaximally non-trivial with respect to ϕ in RQmbC , for some sentence ϕ . Then, for everyformula ψ ( x ) with (at most) a free variable x it holds:(1) [ ∀ xψ ] ∆ = V A ∆ { [ ψ [ x/t ]] ∆ : t ∈ CT er (Ω) } , where V A ∆ denotes an existing infimum inthe Boolean algebra A ∆ ;(2) [ ∃ xψ ] ∆ = W A ∆ { [ ψ [ x/t ]] ∆ : t ∈ CT er (Ω) } , where W A ∆ denotes an existing supremum inthe Boolean algebra A ∆ . roof. (1) By definition, and by the rules from CPL + , [ α ] ∆ ≤ [ β ] ∆ in A ∆ iff ∆ ⊢ RQmbC α → β .Let ψ ( x ) be a formula with (at most) a free variable x . Then [ ∀ xψ ] ∆ ≤ [ ψ [ x/t ]] ∆ forevery t ∈ CT er (Ω), by ( Ax ∀ ). Let β be a formula such that [ β ] ∆ ≤ [ ψ [ x/t ]] ∆ for every t ∈ CT er (Ω). By Remark 11.2 and the definition of order in A ∆ , there is a constant c in C such that [ ψ [ x/c ]] ∆ ≤ [ ∀ xψ ] ∆ . Since [ β ] ∆ ≤ [ ψ [ x/c ]] ∆ , it follows that [ β ] ∆ ≤ [ ∀ xψ ] ∆ . Thisshows that [ ∀ xψ ] ∆ = V A ∆ { [ ψ [ x/t ]] ∆ : t ∈ CT er (Ω) } . Item (2) is proved analogously. ✷ Proposition 11.11
Let ∆ ⊆ F or (Ω) be as in Lemma 11.10. Then, the interpretation map [[ · ]] A ∆ : Sen (Ω U ) → C A ∆ is such that [[ ψ ]] A ∆ = ∗ ([( ψ ) ⊲ ] ∆ ) for every sentence ψ in Sen (Ω U ) .Moreover, [[ ψ ]] A ∆ = 1 ∆ iff ∆ ⊢ RQmbC ( ψ ) ⊲ . In particular, [[ ψ ]] A ∆ = 1 ∆ iff ∆ ⊢ RQmbC ψ forevery ψ ∈ Sen (Ω) . Proof.
The proof is done by induction on the complexity of the sentence ψ in Sen (Ω U ). If ψ = P ( t , . . . , t n ) is atomic then, by using Definition 10.9, the fact that [[ t ]] c A ∆ = ( t ) ⊲ for every t ∈ CT er (Ω U ), and Definition 11.8, we have:[[ ψ ]] A ∆ = P A ∆ ([[ t ]] c A ∆ , . . . , [[ t n ]] c A ∆ ) = P A ∆ (( t ) ⊲ , . . . , ( t n ) ⊲ ) = ∗ ([( ψ ) ⊲ ] ∆ ).If ψ = β for ∈ {¬ , ◦} then, by Definition 10.9 and by induction hypothesis,[[ ψ ]] A ∆ = β ]] A ∆ = ∗ ([( β ) ⊲ ] ∆ )) = ∗ ([( β ) ⊲ ] ∆ ) . If ψ = α β for ∈ {∧ , ∨ , →} , the proof is analogous.If ψ = ∀ xβ then, by Lemma 11.10 and using that U = CT er (Ω), [ ∀ xβ ] ∆ = V A ∆ { [ β [ x/t ]] ∆ : t ∈ U } and so ∗ ([ ∀ xβ ] ∆ ) = V C A ∆ {∗ ([ β [ x/t ]] ∆ ) : t ∈ U } . Then, by Definition 10.9 and byinduction hypothesis:[[ ∀ xβ ]] A ∆ = ^ t ∈ U [[ β [ x/ ¯ t ]]] A ∆ = ^ t ∈ U ∗ ([( β [ x/ ¯ t ]) ⊲ ] ∆ ) = ∗ ([( ∀ xβ ) ⊲ ] ∆ ) . If ψ = ∃ xβ , the proof is analogous to the previous case.This shows that [[ ψ ]] A ∆ = ∗ ([( ψ ) ⊲ ] ∆ ) for every sentence ψ . The rest of the proof followsby Proposition 11.7. ✷ Theorem 11.12 (Completeness of RQmbC w.r.t. BALFI semantics)
For every Γ ∪ { ϕ } ⊆ Sen (Ω) : if Γ | = RQmbC ϕ then Γ ⊢ RQmbC ϕ . Proof.
Suppose that Γ ∪{ ϕ } ⊆ Sen (Ω) is such that Γ RQmbC ϕ . By Proposition 11.4, thereexists a C -Henkin theory ∆ over Ω C in RQmbC , for some nonempty set C of new individualconstants, such that Γ ⊆ ∆ and, in addition, ∆ is maximally non-trivial with respect to ϕ in RQmbC . Consider now B ∆ and A ∆ as in Definitions 11.6 and 11.8, respectively. ByProposition 11.11, [[ ψ ]] A ∆ = 1 ∆ iff ∆ ⊢ C RQmbC ψ , for every ψ in Sen (Ω C ). But then [[ γ ]] A ∆ =1 ∆ for every γ ∈ Γ and [[ ϕ ]] A ∆ = 1 ∆ . Now, let A the reduct of A ∆ to Ω. Hence, A is a structureover B ∆ and Ω such that [[ γ ]] A = 1 ∆ for every γ ∈ Γ but [[ ϕ ]] A = 1 ∆ . From this, = RQmbC ϕ . Inaddition, for every non-empty set { γ , . . . , γ n } ⊆ Γ it is the case that V ni =1 [[ γ i ]] A = 1 [[ ϕ ]] A .Therefore the formula ( γ ∧ ( γ ∧ ( . . . ∧ ( γ n − ∧ γ n ) . . . ))) → ϕ is not valid in ( A , B ∆ ). Thismeans that Γ = RQmbC ϕ . ✷ emark 11.13 The completeness result for
RQmbC w.r.t. BALFI semantics was obtainedjust for sentences, and not for formulas possibly containing free variables (as it was donewith the soundness Theorem 10.15). This can be easily overcome. Recall that the universalclosure of a formula ψ in F or (Ω) , denoted by ( ∀ ) ψ , is defined as follows: if ψ is a sen-tence then ( ∀ ) ψ def = ψ ; and if ψ has exactly the variables x , . . . , x n occurring free then ( ∀ ) ψ def = ( ∀ x ) · · · ( ∀ x n ) ψ . If Γ is a set of formulas in F or (Ω) then ( ∀ )Γ def = { ( ∀ ) ψ : ψ ∈ Γ } . It is easy to show that, for every Γ ∪ { ϕ } ⊆ F or (Ω) : (i) Γ ⊢ RQmbC ϕ iff ( ∀ )Γ ⊢ RQmbC ( ∀ ) ϕ ; and (ii) Γ | = RQmbC ϕ iff ( ∀ )Γ | = RQmbC ( ∀ ) ϕ . From this, a generalcompleteness for RQmbC result follows from Theorem 11.12.
12 Conclusion, and significance of the results
This paper offers a solution for two open problems in the domain of paraconsistency, inparticular connected to algebraization of
LFI s. The quest for the algebraic counterpart ofparaconsistency is more than 50 years old: since the inception of da Costa’s paraconsistentcalculi, algebraic equivalents for such systems have been searched, with different degrees ofsuccess (and failure). Our results suggest that the new concepts and methods proposed inthe present paper, in particular the neighborhood style semantics connected to BALFIs, havea good potential for applications. As suggested in [32], modal logics could alternatively beregarded as the study of a kind of modal-like contradiction-tolerant systems. In alternativeto founding modal semantics in terms of belief, knowledge, tense, etc., modal logic could beregarded as a general ‘theory of opposition’, more akin to the Aristotelian tradition.Applications of paraconsistent logics in computer science, probability and AI, just tomention a few areas, are greatly advanced when more traditional algebraic tools pertain-ing to extensions of Boolean algebras and neighborhood semantics, are used to express theunderlying ideas of paraconsistency. In addition, many logical systems employed in deonticlogic and normative reasoning, where non-normal modal logics and neighborhood semanticsplay an important role, could be extended by means of our approach. Hopefully, our resultsmay unlock new research in this direction. Finally, BALFI semantics for
LFI s opens thepossibility of obtaining new algebraic models for paraconsistent set theory (see [13, 15]) bygeneralizing the well-known Boolean-valued models for ZF (see [5]).
Acknowledgements:
The first and second authors acknowledge support from the NationalCouncil for Scientific and Technological Development (CNPq), Brazil under research grants307376/2018-4 and 306530/2019-8, respectively.
References [1] Avron, A., Non-deterministic matrices and modular semantics of rules. In J.-Y. B´eziau,editor,
Logica Universalis , pages 149–167. Basel: Birkh¨auser Verlag, 2005.[2] Avron, A., Self-Extensional Three-Valued Paraconsistent Logics.
Logica Universalis , 11,297–315, 2017. 303] Avron, A. and B´eziau, J.-Y., Self-extensional three-valued paraconsistent logics have noimplication.
Logic Journal of the IGPL , 25(2):183–194, 2017.[4] Avron, A. and Zamansky, A., A paraconsistent view on B and S5. In: L. Beklemishev,S. Demri and A. M´at´e (eds),
Advances in Modal Logic , volume 11, pages 21–37. London:College Publications, 2016.[5] Bell, J. L.,
Set Theory: Boolean-Valued Models and Independence Proofs , Third edition.Volume 47 of the
Oxford Logic Guides
Series. Oxford University Press, 2005.[6] B´eziau, J.-Y., Idempotent Full Paraconsistent Negations are not Algebraizable.
NotreDame Journal of Formal Logic , 39(1):135–139, 1998.[7] B´eziau, J.-Y., S5 is a paraconsistent logic and so is first-order classical logic. LogicalStudies , 9:301–309, 2002.[8] B´eziau, J.-Y., Paraconsistent logic from a modal viewpoint.
Journal of Applied Logic ,3:7–14, 2005.[9] Blok, W J. and Pigozzi, D.,
Algebraizable Logics . Volume 77(396) of
Memoirs of theAmerican Mathematical Society . Providence: American Mathematical Society, 1989.[10] Bou, F., Esteva, F., Font, J. M., Gil, A., Godo, L., Torrens A. and Verd´u, V., Logicspreserving degrees of truth from varieties of residuated lattices.
Journal of Logic andComputation , 19(6)1031–1069, 2009.[11] Bueno-Soler, J. and Carnielli W., Experimenting with Consistency. In: V. Markin andD. Zaitsev (eds.),
The Logical Legacy of Nikolai Vasiliev and Modern Logic , pages 199–221. Synthese Library (Studies in Epistemology, Logic, Methodology, and Philosophy ofScience), vol 387. Cham: Springer, 2017.[12] Carnielli, W. A. and Coniglio, M. E., Combining Logics. In: E.N.Zalta (ed.),
The Stanford Encyclopedia of Philosophy , 2007. URL:http://plato.stanford.edu/entries/logic-combining/[13] Carnielli, W. A. and Coniglio, M. E., Paraconsistent set theory by predicating on con-sistency.
Journal of Logic and Computation
Paraconsistent Logic: Consistency, Contradictionand Negation . Volume 40 of
Logic, Epistemology, and the Unity of Science series. Cham:Springer, 2016.[15] Carnielli, W. A. and Coniglio, M. E., Twist-Valued Models for Three-valued Paracon-sistent Set Theory. arXiv :1911.11833 [math.LO], 2019.[16] Carnielli, W. A., Coniglio, M. E., Gabbay, D., Gouveira, P. and Sernadas, C.,
Analysisand Synthesis of Logics: How to Cut and Paste Reasoning Systems . Volume 35 of
AppliedLogic series. Dordrecht: Springer, 2008. 3117] Carnielli, W. A., Coniglio, M. E. and Marcos, J., Logics of Formal Inconsistency. In: D.M. Gabbay and F. Guenthner, editors,
Handbook of Philosophical Logic (2nd. edition) ,volume 14, pages 1–93. Dordrecht: Springer, 2007.[18] Carnielli, W. A., Coniglio, M. E., Podiacki, R. and Rodrigues, T., On the way to a widermodel theory: Completeness theorems for first-order logics of formal inconsistency.
TheReview of Symbolic Logic , 7(3):548–578, 2014.[19] Carnielli, W. A. and de Alcantara, L. P., Paraconsistent Algebras.
Studia Logica
Advances in Contemporary Logicand Computer Science , pages 33-52. Volume 235 of
Contemporary Mathematics Series .Providence: American Mathematical Society, 1999.[21] Carnielli, W. A. and Marcos, J., A taxonomy of C-systems. In: W. A. Carnielli, M. E.Coniglio, and I. M. L. D’Ottaviano, editors,
Paraconsistency: The Logical Way to theInconsistent , volume 228 of
Lecture Notes in Pure and Applied Mathematics , pages 1–94.New York: Marcel Dekker, 2002.[22] Chellas, B.F.,
Modal Logic: An Introduction . New York: Cambridge University Press,1980.[23] Coniglio, M. E., Esteva, F. and Godo, L., Logics of formal inconsistency arising fromsystems of fuzzy logic.
Logic Journal of the IGPL , 22(6):880–904, 2014.[24] Coniglio, M.E., Figallo-Orellano A. and Golzio, A. C., First-order swap structures se-mantics for some Logics of Formal Inconsistency.
Journal of Logic and Computation , toappear, 2020. Preprint available at arXiv :1912.10277 [math.LO].[25] Coniglio, M.E. and Prieto-Sanabria, L., Modal logic S4 as a paraconsistent logic witha topological semantics. In: P. Gouveia, C. Caleiro and F. Donisio (eds.),
Logic andComputation: Essays in Honour of Amilcar Sernadas , pages 171–196. Volume 33 ofTributes Series. London: College Publications, 2017.[26] da Costa. N. C. A.,
Sistemas formais inconsistentes ( Inconsistent formal systems , inPortuguese). Habilitation thesis, Universidade Federal do Paran´a, Curitiba, Brazil, 1963.Republished by Editora UFPR, Curitiba, Brazil,1993.[27] de Ara´ujo, A. L., Alves, E. H. and Guerzoni, J. A. D., Some relations between modal andparaconsistent logic.
The Journal of Non-Classical Logic , 4(2):33–44, 1987. Available at .[28] Ertola, R., Esteva, F., Flaminio, T, Godo, L. and and Noguera, C., Exploring paracon-sistency in degree-preserving fuzzy logics.
Soft Computing , 19(3):531–546, 2015.[29] Givant, S. and Halmos, P.,
Introduction to Boolean Algebras . New York: Springer, 2009.3230] J´onsson, B., A Survey of Boolean Algebras with Operators. In: I.G. Rosenberg andG. Sabidussi, editors,
Algebras and Orders , volume 389 of NATO ASI Series (Series C:Mathematical and Physical Sciences), pages 239–286. Dordrecht: Springer, 1993.[31] Lewin, R. A., Mikenberg, I. F. and Schwarze, M. G., C is not algebraizable. Notre DameJournal of Formal Logic , 32(4):609–611, 1991.[32] Marcos, J., Modality and paraconsistency. In M. Bilkova and L. Behounek, editors. TheLogica Yearbook 2004,
Proceedings of the XVIII International Symposium promoted bythe Institute of Philosophy of the Academy of Sciences of the Czech Republic . Filosofia,Prague, 2005.[33] Marcos, J., Nearly every normal modal logic is paranormal.
Logique et Analyse , 48(189-192):279–300, 2005.[34] Mortensen, C., Every quotient algebra for C is trivial. Notre Dame Journal of FormalLogic , 21(4):694–700, 1980.[35] Mortensen, C., Paraconsistency and C . In G. Priest and R. Routley and J. Nor-man (eds), Paraconsistent Logic: Essays on the Inconsistent , pages 289–305. Munich:Philosophia Verlag, 1989.[36] Nipkow, T., Paulson, L. C. and Wenzel, M.,
Isabelle/HOL: a proof assistant for higher-order logic . Volume 2283 of
Springer Science & Business Media . Springer, 2002.[37] Pacuit, E.,
Neighborhood semantics for modal logic . Springer, 2017.[38] Segerberg, K.,
Classical Propositional Operators: An exercise in the foundations of logic .Volume 5 of
Oxford Logic Guides . Oxford: Clarendon Press, 1982[39] Seoane, J. and de Alcantara, L. P., On da Costa algebras.
The Journal of Non-ClassicalLogic
Theory of logical calculi: Basic theory of consequence operations . Volume199 of
Synthese Library . Dordrecht: Reidel, 1988.[41] Zanardo, A., Sernadas, A. and Sernadas, C., Fibring: Completeness preservation.