A Tableau Calculus for Dummett Logic Based on Increasing the Formulas Equivalent to the True and the Replacement Rule

Transcription

1 A Tableau Calculus for Dummett Logic Based on Increasing the Formulas Equivalent to the True and the Replacement Rule Guido Fiorino 1 Dipartimento di Metodi Quantitativi per le cienze Economiche ed Aziendali, Università di Milano-Bicocca, Piazza dell Ateneo Nuovo, 1, Milano, Italy. guido.fiorino@unimib.it Abstract. In this paper we present a multiple premise tableau calculus for propositional Dummett Logic. The aim is to improve the efficiency of the known decision procedures. The investigation is related to the possibility of reducing the size of the deduction by the replacement rule and rules that introduce formulas equivalent to the logical constant True. In this way we want to bound the application of the multiple premise rule, which is the main source of inefficiency, and improve the practical performances of the decision procedure. The timings of a prolog implementation are provided. 1 Introduction The aim of this paper is to present a new tableau calculus for Dummett logic to be used as a core of a decision procedure faster than the known implementations. The history of this logic starts with Gödel, who studied the family of logics semantically characterizable by a sequence of n-valued (n > 2) matrices ([9]). In paper [5] Dummett studied the logic semantically characterized by an infinite valued matrix which is included in the family of logics studied by Gödel and proved that such a logic is axiomatizable by adding to any Hilbert system for propositional intuitionistic logic the axiom scheme (p q) (q p). Moreover, it is well-known that such a logic is semantically characterizable by linearly ordered Kripke models. Dummett logic has been extensively studied also in recent years for its relationships with computer science ([3]) and fuzzy logics ([11]). To perform automated deduction both tableau and sequent calculi have been proposed. Paper [1] provides tableau calculi whose distinguishing feature is a multiple premise rule for implicative formulas signed with F. We recall that the sign F comes from mullyan ([16, 8]) and labels those formulas that in a sequent calculus occur in the right-hand side of (as it is explained in ection 2, the sign F has a meaning also in terms of Kripke semantics). A tableau calculus derived from those of [1] is provided in paper [6]. Its main feature is that the depth of every deduction is linearly bounded in the length of the formula to be proved.

2 The approach of [1], based on characterizing Dummett logic by means of the multiple premise rule, has been criticized because, from the worst case analysis perspective, there are simple examples of sets of formulas giving rise to a factorial number of branches in the number of formulas in the set. Paper [4] shows how to get rid of the multiple premise rule. New rules are provided whose correctness is strictly related to the semantics of Dummett logic. These ideas have been further developed in [12, 13], and in paper [14] a graph-theoretic decision procedure is described and implemented. The approach introduced in [4] has also disadvantages with respect to the multiple premise rule proposed in [1] and these disadvantages have been considered in [7], where also a new version of the multiple premise rule is proposed. This version, from a practical point of view, can reduce the branching when compared with the original one. Paper [7] also provides an implementation that outperforms the one of [14], thus proving that the approach based on the multiple premise rule of [1] deserves attention also from the practical point of view. As a matter of fact, on the one hand the rules of [4] give rise to two branches at most, on the other hand there are cases of formulas that multiple premise calculi decide with a number of steps lower than the calculi based on [4]. In this paper we continue our investigation around multiple premise calculi for Dummett logic. The tool we use to prove our results is the characterization of Dummett logic via linearly ordered Kripke models, whose elements are considered worlds or states of knowledge ordered w.r.t. and α β means that, roughly speaking, β is in a subsequent point of the time w.r.t. α. With the aim to reduce the size of the proofs, we present a calculus whose main feature is handling two kinds of signed formulas: the first kind is used to describe the semantical status of a formula at the present state of knowledge, the second kind is used to describe the semantical status of a formula to the next state of knowledge. The introduction of the signs related to the future state of knowledge allows to introduce specialized rules to draw deductions also about the present state of knowledge. In particular, we refer to the introduction of specialized and single conclusion rules to handle the formulas of the kind F(A B) (to be read at the present state of knowledge, the fact A B is not known ), thus reducing the branching generated by the multiple premise rule. These two semantical levels allow to conceive rules such that when from a premise related to the present, facts about the future are deduced, among these facts there is at least one that at the next state of knowledge becomes explicitly known, being explicitly unknown at the present state of knowledge. This implies that moving from the present to the future new facts becomes known. This has also a correspondence in the construction of the Kripke counter model, where there are no two elements exactly forcing the same propositional variables. uch an increment of known information, that is formulas equivalent to the truth ( ) is exploited by the replacement rule, which replaces all the occurrences of a formula proved to be equivalent to the with. uch a replacement reduces the size of the set to be decided. Further reductions are possible by applying replacements based on the truth table of the connectives.

3 Rules based on the replacement have been proved effective both in classical and intuitionistic logic ([15, 2]), thus we have developed a calculus whose rules are designed to conclude as much as possible about formulas equivalent to. We remark that by using the appropriate data structures to represent formulas and sets of formulas, the implementation of the replacement rule and the reduction rules based on the truth table of the connectives can be performed in constant time. The preliminary results of our prolog prototype show that the calculus we provide is suitable for an implementation outperforming those quoted above. 2 Basic definitions, the calculus and general considerations We consider the propositional language based on a denumerable set of propositional variables PV, the boolean constants and and the logical connectives,,,. We call atoms the elements of PV {, }. In the following, formulas (respectively set of formulas and propositional variables) are denoted by letters A, B, C... (respectively, T, U,... and p, q, r,... ) possibly with subscripts or superscripts. From the introduction we recall that Dummett Logic (Dum) can be axiomatized by adding to any axiom system for propositional intuitionistic logic the axiom scheme (p q) (q p) and a well-known semantical characterization of Dum is by linearly ordered Kripke models. In the paper model means a linearly ordered Kripke model, namely a structure K = P,,ρ,, where P,, ρ is a linearly ordered set with ρ minimum with respect to and is the forcing relation, a binary relation on P (PV {, }) such that: (i) if α p and α β, then β p; (ii) for every α P, α holds and α does not hold. Hereafter we denote the members of P by means of lowercase letters of the Greek alphabet. The forcing relation is extended in a standard way to arbitrary formulas of our language as follows: 1. α A B iff α A and α B; 2. α A B iff α A or α B; 3. α A B iff, for every β P such that α β, β A implies β B; 4. α A iff for every β P such that α β, β A does not hold. We write α A when α A does not hold. It is easy to prove that for every formula A the persistence property holds: If α A and α β, then β A. We say that β is immediate successor of α iff α < β and there is no γ P such that α < γ < β. A formula A is valid in a model K = P,,ρ, if and only if ρ A. It is well-known that Dum coincides with the set of formulas valid in all models. The rules of our calculus D for Dum are in Figures 1 and 2. D works on signed formulas, that is well-formed formulas prefixed with one of the signs {T, F, F u, F n, F c, T cl, T n }, and on sets of signed formulas (hereafter we omit the word signed in front of formula in all the contexts where no confusion

4 arises). Before to give the intuition behind the rules of the calculus the meaning of the signs is provided by the relation realizability ( ) defined as follows: Let K = P,,ρ, be a model, let α P, let H be a signed formula and let be a set of signed formulas. We say that α realizes H (respectively α realizes and K realizes ), and we write α H (respectively α and K ), if the following conditions hold: 1. α TA iff α A; 2. α F c A iff α A; 3. α T cl A iff α A; 4. α FA iff α A; 5. α F u A iff α FA and for every β > α, β TA; 6. α F n A iff there exists β > α, β FA; 7. α T n A iff for every β > α, β TA; 8. α iff α realizes every formula in ; From the meaning of the signs we get the conditions that make a set of formulas inconsistent. A set is inconsistent if one of the following conditions holds: -T ; -F ; -F u ; -F c ; -{F u A, F n B}; -T cl ; -F n ; -{T n, F n A} ; -{F u, F n A} ; It is easy to prove the following Proposition 1. If a set of formulas is inconsistent, then for every Kripke model K = P,,ρ, and for every α P, α. Proof. We consider some significant cases. Let T cl. We show that α. Let us suppose that α, then α. By definition of forcing of negation, there exists β P such that β. ince no world forces it follows that for every α P, α. Let {F u A, F n B}. By absurd, let us suppose that α, then α F u A and α F n B. ince α F n B, there exists β P such that α < β. By definition of F u, it follows that α A and for every β P, if α < β, then β A. Note that such a β exists. ince K is a linearly ordered Kripke model, by definition of negation it follows that α A. Thus we have that α A and α A, absurd. The other cases are easy to prove. A proof table (or proof tree) for is a tree, rooted in and obtained by the subsequent instantiation of the rules of the calculus. The premise of the rules are instantiated in a duplication-free style: in the application of the rules we always consider that the formulas in evidence in the premise are not in. We say that a rule R applies to a set U when it is possible to instantiate the premise of R with the set U and we say that a rule R applies to a formula H U (respectively the set {H 1,..., H n } U) to mean that it is possible to instantiate the premise of R taking as U \ {H} (respectively U \ {H 1,..., H n }).

5 , T(A B), F(A B), F c(a B), T cl(a B), TA, TB T, FA, FB F, F ca, F Fc cb, T cl A, T cl B T cl, T(A B), F(A B), F c(a B), T cl(a B), TA, TB T, FA, FB F, F ca, F cb Fc, T cl A, T cl B T cl, T( A), F( A), F c( A), T cl( A), F ca T, T cl A F, T cl A Fc, F ca, F c(a B), T cl(a B), T cl A, F cb Fc, F ca, T cl B T cl T cl, T((A B) C), T( A B), T((A B) C), T(A (B C)) T, T cl A, TB T, T(A C), T(B C) T, T((A B) C) T with p a new atom, F(A p), T(p C), T(B p), TC, F u(a B), F(A B), FA, F ub, T nb, FB, F ua, T Fu na, TA, F ub, T nb, F n(a B) F, T n(a B), T n( A), T na, T nb Tn, F ca Tn, F(A B), TA, F F 1, provided TnB ub, F u(a B), TA, F ub, T nb Fu, F u(a B), F ua, T na, T nb, TA, T nb, F ub Fu, T(A B) cl, F ca cl, TB T -cl, T cla cl, TA T cl-cl provided only contains T,F c, T cl and T n-formulas, F u( A), F u cl, TA Fu cl F u,, T n cl T n, where cl = {A A, {T, F c}} {F ca FA } {TA T cl A } {F ca F ua } Fig. 1. The invertible rules of D. A closed proof table is a proof table whose leaves are all inconsistent sets. A closed proof table is a proof of the calculus and a formula A is provable iff there exists a closed proof table for {F u A}. We refer to [10] for full details on tableaux systems. The calculus contains two invertible rules. In ection 4 we prove that it is possible to devise a complete strategy that relies on respecting a particular sequence in the application of the rules: T cl -Atom is applied if no other rule is applicable and F n is applied if no other rule but T cl -Atom is applicable.

6 , T cl A cl, TA φ, TA T cl-atom, provided does not contain F n-formulas. where φ = {TA TA } {F ca F ca } {TA T cl A } {TA F ua } and cl = {A A, {T, F c}} {F ca FA } {TA T cl A } {F ca F ua };, F na 1,..., F na u... c, F na 1,..., F na j 1, F ua j, FA j+1,..., FA u... where c = {A A and {T, F c, T cl }} {TA F ua } {TA T na }; F n Fig. 2. The non-invertible rules of D., TA [A/ ], TA Replace T, F ca [A/ ], F ca Replace Fc, T cl A [ A/ ], T cl A Replace T cl, F na, F ub, F na[b/ ], F ub, A, T nb Replace Tn, with {Tn, Fn}, A[B/ ], T nb Fig. 3. The Replacement rules Replace Fu imp [A / ] imp [A /A] imp [ A/ ] imp [ / ] imp [ A/ ] imp [ A/A] imp [A / A] imp [ / ] imp [A /A] imp [A / ] imp [ A/A] imp [ A/A] imp [ A/ ] imp [A / ] Fig. 4. The implification rules To justify the introduction the signs F u, F n and T n in the object language, we start from a motivating case. Let us suppose that a world α of a model K realizes F = {F(A 1 B 1 ),..., F(A n B n )}. Then there exists the last element β α such that β TA j, FB j, \ {F(A j B j )}. Without loss of generality we let j = 1. Analogously, there exists an element γ >= β such that γ TA i, FB i, TA 1, \ {F(A 1 B 1 ), F(A i B i )}. Without loss of generality we let i = 2. Now, β < γ or β = γ. If β < γ, then since β is last element where FB 1 holds, it follows that γ TB 1 holds. If β = γ, then β is the last element

7 where both B 1 and B 2 are not forced. The example shows that we can give a rule (F ) taking into account that if α F(A i B i ), for i = 1,..., n, then the set {TA i, FB i } is realized in θ {α, β} and for every world γ > θ, γ TB i holds. The sign F u aims to codify such a semantical property of B i. Generalizing the case given above, let F = {F(A 1 B 1 ),..., F(A n B n )} and let be a set of formulas. Let us suppose that α, F. Then for every F(A i B i ) F let us consider the element β i such that β i TA i, FB i and for every γ > β i, γ TA i, TB i. That is β i is the maximum element such that β i TA i, FB i. ince α F(A i B i ), such an element β i exists. Let β j = min{β 1,..., β n }, with j {1,..., n}. There are two cases: (i) β j = α, than we conclude that β realizes the set, TA j, F u B j, F \ {F(A j B j )}; (ii) if β j > α, then β j realizes the set c, TA j, F u B j, F \ {F(A j B j )}, where, by the meaning of the signs, c consists of: (i) the formulas of signed with T, F c and T cl ; (ii) the formulas TA such that F u A. With the aim to reduce the size of the proofs, we add further considerations justifying the introduction of the sign F n. Let us suppose that beside α, F we also know that for every model β 1,..., β j 1 > β j holds. This rules out that β j TA i, F u B i, for i = 1,..., j 1, and implies that there exists an element γ such that β j < γ and γ F(A i B i ). The sign F n aims to codify that the formulas F(A i B i ) have the following semantical property: there exists an element β j < γ such that γ F(A i B i ), thus the element β j is not the maximum element realizing F(A i B i ). By the meaning of F n it follows that β j c, F n (A 1 B 1 ),..., F n (A j 1 B j 1 ), TA j, FB j, F \ {F(A 1 B 1 ),..., F(A j 1 B j 1 )}, where by the meaning of the signs, c consists of: (i) the formulas of signed with T, F c and T cl ; (ii) the formulas TA such that F u A. These are the intuitions justifying the rules F n and F. In particular the aim of F is to distinguish if the last element realizing F(A B) is the same element α realizing the premise or an element β such that α < β. The rule F n consider the formulas of the kind F n A. For every formula there exist a world β such that α < β and β is the last element such that β A. The minimum of such β does not force any formula in evidence in the premise. The correctness of the rule F n can be explained as follows. Let us suppose that α, F n A 1,..., F n A u. Thus there exists β i such that α < β i and β i F u A i, for i = 1,..., u. We notice that β i realizes all the T, F c and T cl formulas in and β i TC if F u C. Moreover, if β 1 = min{β 1,..., β u }, then β 1 F u A 1, FA 2,..., FA u. If β 2 is the minimum and we know that there is no model realizing α, F n A 1,..., F n A u having β 1 as the minimum, then, since β 2 < β 1 holds, we conclude that β 2 F n A 1, F u A 2, FA 3,..., FA u. Analogously, if β i is the minimum and we know that there is no model realizing α, F n A 1,..., F n A u having β 1,... or β i 1 as the minimum, then, since β i < β 1,..., β i 1 holds, we conclude that β i F n A 1,..., F n A i 1, F u A i, FA i+1,..., FA u. The sign T n is strictly related to F u and conveys redundant information with respect to the completeness. For this reason we give only two rules to treat the formulas of this kind and the replacement rules. This presentation of the calculus takes into account of the implementation of the prolog prototype.

8 We have decided to embed in the logical part the information conveyed from the F u -formulas. The explicit introduction of T n emphasize how to handle the information of the F u -formulas and its computational cost. An alternative presentation could avoid to insert the sign T n. In this case the rule F 1 becomes, F(A B), F u B, TA, F u B Also the replacement rules for T n can be restated as rules involving F u. The rules T n and T n deserve some considerations. From the point of view of the presentation of the logical calculus they are not relevant and can be deleted from Figure 1. From the practical point of view, the implementation of these rules is useful because they break down the information deriving from F u formulas and can be used, as an example to turn a F -formula into a F n -formula. Thus one wonders how to insert them in the decision procedure. Our implementation treats explicitly the derived from F u -formulas by means of the sign T n and of the logical rules. A different choice could be to use the F u -formulas and implicitly break down the semantical information of the F u -formulas to decide, as an example, if a given F -formula can be turned into a F u -formula, or if F n A is equivalent to (the inconsistent) formula F n. Our choice is to insert all the machinery in the logical apparatus of the calculus. 3 Correctness To prove the correctness of D with respect to Dummett logic we need to prove that, if there exists a closed proof table for {F u A}, then A is a valid formula in Dummett logic. The main step is to prove that the rules of the calculus preserve realizability: Proposition 2. For every rule of D, if a model realizes the premise, then there exists a model realizing at least one of the conclusions. Proof. Let α be an element of K = P,,ρ,. We analyze the correctness of some rules of D. Rule F u : if α, F u (A B), then, by definition of F u, α A B and for every β P, if α < β, then β A B. This implies, by definition of forcing of an implicative formula, that α A and α B hold. By the preservation of forcing we get that, for every β P, if α < β, then β A, thus β B. Therefore α F u B. Rule F u : if α, F u (A B) then, by definition of F u, α A B, thus α A or α B. Moreover, for every β P, if α < β, then β A B, thus β A and β B. We have two possible cases: (i) α A, thus α F u A, α T n A and α T n B; (ii) α A, thus α TA, α F u B and α T n B.

9 Rule F u : if α, F u ( A), then we notice that no F n -formula is in. By definition of F u, α A, thus there exists γ P such that α γ and γ A. Moreover, for every β P, if α < β, then β A. By definition of forcing, an element cannot force a formula and its negation, thus the only possibility is that α is the final element of K, thus γ = α. ince α is the final element of K, for every formula B, α B implies α B and α B implies α B. Thus α cl. Rule T cl -Atom: if α, T cl A, then we have two cases: (i) α is not the final element φ. By the semantical meaning of the signs, it follows that φ realizes all the T and F c -formulas in and if B, with {T cl, F u }, then φ TB; (ii) if α is the final element then it is immediate to check that α realizes the leftmost conclusion of T cl -Atom. Rule T n : if α, T n, then for every β P, if α < β, then β. ince no world forces the only possibility is that α < β does not hold, thus α is the final element of K. ince α behaves as a classical model with respect to the semantics of the connectives, we get that α cl. Theorem 1. If there exists a proof table for A, then A is valid in Dum. Proof. By hypothesis there exists a proof table starting from {F u A} whose leaves are all inconsistent. ince an inconsistent set is not realizable and the rules of D preserve the realizability, it follows that F u A is not realizable, hence every element of every model forces A and we conclude that A is valid. The aim of the sign F u is to label those formulas that after an application of F n become equivalent to. After an application of the rule F n, the formulas signed with F u become signed with T, thus they are equivalent to. Hence the rule Replace T of Fig. 3 are applicable. The occurrences of are removed by using the simplification rules of Fig. 4. We refer to the whole machinery of Fig. 3 and 4 as simplification rules. Its an easy task to check that the rules are invertible. We emphasize that Replace F c and Replace T cl are a specialization of Replace T obtained by exploiting the meaning of F c and T cl. The rule Replace F u exploits the meaning of F u and F n to perform a replacement: if α F n A, F u B holds, then we conclude β FA, TB, with β immediate successor of α. Thus β FA[B/ ] and α FA[B/ ] hold. It can be noticed that in the conclusion of the F u -rules there is a duplication of a subformula of the premise. This duplication, that occurs with sign T n, is correct but not necessary to the completeness and is related to the logical treatment of the information under sign F u. It has to be noticed that this duplication does not explode in an exponential number of formulas since the presence of the rule Replace T n allows to treat once the T n -formulas generated by duplication from F u -formulas. As an example, consider F u and let us suppose that in B occurs as subformula of T n C in. By applying Replace T n the occurrence of B in C is replaced by, thus the rules of the calculus handle the connectives of B twice: once when F u B is handled and once when T n B is handled. We also remark that there are only two rules for T n -formulas. This implies that the overhead

10 necessary to manage by T n -formulas the information conveyed by F u -formulas does not implies the generation of branches and requires at most a number of applications of rules of D which is linear in the length of the F u -formulas. Compared with the calculus in paper [7], introducing the sign F u has a price: proof tables can be wider, because the combined action of F and F n. To reduce this problem there is the rule F 1, which is not necessary to the completeness and represents an example of exploitation of T n -formulas. Moreover, since the application of rules in Fig.2 turns the F u -formulas into T-formulas, that is formulas equivalent to the logical value, we investigate if this feature can reduce the depth of the proofs. We have experienced that simplification rules reduce the size of the proofs, for this reason we are devising a calculus that introduces as much as possible formulas equivalent to the logical value. 4 A trategy to Decide Dummett Logic and Its Completeness In the following we sketch the recursive procedure Dum(). Given a set of formulas, Dum() returns either a closed proof table for or NULL (if there exists a model realizing ). To describe Dum we use the following definitions and notations. We call α-rules and β-rules the rules of Figure 1 with one conclusion and with two conclusions, respectively. The α-formulas and β-formulas are the kind of the signed formulas in evidence in the premise of the α-rules and β-rules, respectively. Let be a set of formulas, let H be an α or β-formula. With Rule(H) we denote the rule corresponding to H in Figure 1. Let 1 or 1 2 be the nodes of the proof tree obtained by applying to the rule Rule(H). If T ab 1 T ab 1 Rule(H) and T ab 2 are closed proof tables for 1 and 2 respectively, then or T ab 1 T ab Rule(H) 2 denote the closed proof table for defined in the obvious way. Moreover, for H different from F u and T n, R i (H) (i = 1, 2) denotes the set containing the formulas of i which replaces H. For instance: R 1 (T(A B)) = { TA, TB }, R 1 (T(A B)) = {TA}, R 2 (T(A B)) = {TB}, In the case of F n we generalize the above notation. Let Fn be the set of all the F n -formulas of. Let 1... n be the nodes of the proof tree obtained by applying to the rule F n. If T ab 1..., T ab n are closed proof tables for 1,..., n, respectively, then T ab 1... T ab 2 F n is the closed proof table for. With R i ( Fn ) we denote the set of formulas that replaces the set Fn in the i-th conclusion of F n. For example, given Fn = {F n A 1, F n A 2, F n A 3 }, R 2 ( Fn ) = {F na 1, F u A 2, FA 3 }. We consider implification as function that applies to its parameters the rules of Figures 3 and 4 as long as possible. Function Dum () 1. If is an inconsistent set, then Dum returns the proof ; 2. Let =implification(). If, then let π = Dum( ). If π is a proof, then Dum returns π imp, otherwise Dum returns NULL;

11 3. If the rule F u or T n applies to, then let H be the formula T n or F u. If Dum(( \ {H}) cl ) returns a proof π, then Dum returns the proof Rule(H), otherwise Dum returns NULL; π 4. If the rule T cl -cl or F u applies to, then let H be a formula of the kind T cl A or F u. If Dum(( \ {H}) cl {TA}) returns a proof π, then Dum returns the proof π Rule(H), otherwise Dum returns NULL; 5. If an α-rule applies to, then let H be a α-formula of. If Dum(( \ {H}) R 1 (H)) returns a proof π, then Dum returns the proof π Rule(H), otherwise Dum returns NULL; 6. If the rule T -cl applies to, then let H be a formula of the kind T(A B). Let π 1 = Dum(( \ {H}) cl {F c A}) and π 2 = Dum(( \ {H}) cl {TB}). If π 1 or π 2 is NULL, then Dum returns NULL, otherwise Dum returns π 2 T -cl; 7. If a β-rule applies to, then let H be a β-formula of. Let π 1 = Dum(( \ {H}) R 1 (H)) and π 2 = Dum(( \ {H}) R 2 (H)). If π 1 or π 2 is NULL, then Dum returns NULL, otherwise Dum returns π 1 π 2 Rule(H); 8. If the rule F n applies to, then let Fn = {F n A } and n = Fn. If there exists i {1,..., n}, such that π i = Dum(( \ Fn ) c R i ( Fn )) is NULL, then Dum returns NULL. Otherwise π 1,..., π n are proofs and Dum returns π 1... π n F n ; 9. If the rule T cl -Atom applies to, then let H be a T cl -Atom formula of. Let π 1 = Dum(( \ {H}) cl R 1 (H)) and π 2 = Dum(( \ {H}) φ R 2 (H)). If π 1 or π 2 is NULL, then Dum returns NULL, otherwise Dum returns 10. If none of the previous points apply, then Dum returns NULL. end function Dum. π 1 π 1 π 2 Rule(H); We emphasize that function Dum respects a particular sequence in the application of the rules: T cl -Atom is applied if no other rule is applicable and F n is applied if no other rule but T cl -Atom is applicable. As a result no backtracking step is necessary. By inspecting the rules of the calculus it follows that the procedure terminates. In particular the rightmost conclusion of the rule F only change the sign of the formula in evidence in the premise, but we remark that the rule F n handles these kind of formulas, thus the combined action of F and F n allows to prove that there is no an infinite loop handling F -formulas. In order to get the completeness of Dum, in the following it is proved that given a set of formulas, if the call of Dum() returns NULL, then there is enough information to build a model K = P,,ρ, such that ρ. Lemma 1 (Completeness). Let be a set of formulas and suppose that Dum() returns NULL. Then, there exists a Kripke model K = P,,ρ, such that ρ. Proof. By induction on the number of nested recursive calls.

12 Basis: There are no recursive calls. Then tep 10 has been performed. We notice that is not inconsistent (otherwise tep 1 would have been performed) and does not contain neither F n -formulas nor T cl -formulas (otherwise tep 8 or tep 9 would have been performed). Indeed, only contains atomic formulas signed with T,F,F c, T n, F u and formulas of the kind T(p A) with p atomic and Tp (otherwise, if Tp holds, then tep 2 applies and at least a recursive call to Dum would have been performed). It is easy to prove that the model K = P,,ρ,, where P = {ρ}, ρ ρ and ρ p iff Tp, realizes. tep: By induction hypothesis we assume that the proposition holds for all sets such that Dum( ) requires less than n recursive calls. We prove the proposition holds for a set such that Dum() requires n recursive calls by inspecting all the possible cases where the procedure returns the NULL value. NULL value returned performing tep 4 with H of the kind F u ( A). By induction hypothesis there exists a model K realizing cl, TA. Notice that does not contain F n -formulas, otherwise would be inconsistent, and cl contains only T and F c -formulas. This implies that all the formulas occurring in the subsequent recursive calls are signed with T and F c. Thus no step related to the rules of Fig. 2 is employed. We conclude that ρ is the only element of K and ρ cl implies ρ. Moreover, ρ maximal element of K and ρ TA imply ρ F u ( A). NULL value returned performing tep 8. We notice that is consistent, thus no formula of the kind F u ( A) belongs to. The set contains atomic formulas and formulas of the kind T(p A), F n (A B), F n (A B), F n (A B). ince the NULL instruction in tep 8 has been performed, at least a π i is NULL, then by induction hypothesis there is a model K = P,, ρ, realizing ( \ Fn ) c R i ( Fn ). We define a Kripke model K = P {ρ},, ρ,, where P {ρ} =, = {(ρ, α) α P } = {(ρ, p) Tp } ince K is a Dummett model realizing (\ Fn ) c, it follows that K is a Dummett model. As a matter of fact, ρ is the immediate successor of ρ and TA implies TA ( \ Fn ) c, thus the forcing relation is preserved. Moreover, F u A implies TA ( \ Fn ) c and the consistency of implies TA. Finally, A, with {F c, T cl }, implies A ( \ Fn ) c, thus we have proved that K ( \ Fn ) c R i ( Fn ). It is an easy task to check that K realizes. NULL value returned performing tep 9. We notice that when tep 9 is reached does not contain any F n -formula. Thus if a T cl -formula belongs to the recursive call is performed and it is a correct application of the rule T cl -Atom. ince tep 9 returns NULL one between π 1 or π 2 is NULL. If π 1 is NULL, then by induction hypothesis there is a model K = P,,ρ, such that ρ ( \

13 Formula EPDL New-EPDL YJ YJ YJ YJ YJ YJ YJ YJ YJ YJ YJ YJ Formula EPDL New-EPDL YJ YJ YJ YJ YJ YJ YJ YJ YJ YJ YJ YJ YJ YJ YJ YJ Fig. 5. EPDL and New-EPDL on ILTP formulas. {T cl A}) cl TA. By the meaning of ( \ {T cl A}) cl immediately we get that ρ. If π 2 is NULL, then, by induction hypothesis, the recursive call returns a model K = P,, ρ, realizing (\{T cl A}) φ {TA}. We define a model K = P {ρ},, ρ, as in the above point. Now, ρ TA implies ρ T cl A. Moreover, by definition of ( \ {T cl A}) φ, it is immediate to prove that ρ ( \ {T cl A}) holds. 5 Experimental Results The prolog prototype EPDL-new 1 implements the strategy described in ection 4. In Fig. 5 we report the comparisons on the formulas of ILTP library. On the missing families the behaviour of the implementations is similar. On the the families of formulas YJ201, YJ205, YJ207 and YJ208 EPDL-new is clearly faster than EPDL and the timings of EPDL-new grow slower than EPDL. As regard the family YJ202 (the pigeon principle) EPDL-new is clearly slower. This is due to the overhead to manage the T n -formulas introduced in the deduction that do not give any advantage to decide this formula. The overhead to manage the T n -formulas is twofold. In the deduction occur formulas of the kind T n (A B), thus the decision procedure applies the rule T n. Moreover, the application of the rules in Fig. 3 and 4 requires linear time in the number of connectives and atoms in the set of formulas. ince the set managed by EPDL-new 1 available from guidofiorino/epdl.jsp

14 is bigger than EPDL we have such an increment in the timings. As we noticed in the paper, by employing the data structures of [2] the time required to implement the rules Fig. 3 and 4 can be reduced to constant time, thus reducing the impact of the overhead. As regard the family formulas YJ211, we notice that the different timings have to be charged to the implementation of rules in Fig. 3 and 4. As a matter of fact to decide the last formula of the family EPDL takes 127 rules and EPDL-new 147, where the difference in the number of the rules is due to the application of T n -rule. We also remark that the growing ratio of the two implementations is similar. Finally EPDL decides YJ by applying 54 rules whereas EPDL-new applies 81. These formulas are huge (YJ is a formula in 14 variables, containing connectives and variable occurrences). Because of the duplication of the calculus, the sets of formulas managed by EPDL-new contains more symbols that those managed by EPDL and the difference in timings is mostly chargeable to the implementation of the rules in Fig. 3 and 4 than the difference on the number of applications of rules. Another clue to support our statement is that EPDL-new requires to apply 63, 69, 75 and 81 rules to decide respectively YJ , YJ , YJ and YJ , whereas to decide the same formulas EPDL applies respectively 42, 46, 50 and 54 rules. We remark that the number of rules increases of a constant value for both provers, but the timings of the EPDL and EPDL-new approximately double. ummarizing, on some families formulas the ideas on which the calculus is based on do not apply, as a result the deduction is slowed-down. This difference is remarkable on the family YJ202, the pigeon principle, that, apart from the first rule, is decided by a deduction requiring rules for connectives and. We point out that on the families YJ202, YJ211 and YJ212 the multiple premise rule F n is not applied. We have argued that implementing the calculus with better data structures, on the family formulas YJ211 and YJ212 the difference in timings between EPDL and EPDL-new can be reduced almost to zero. Finally we remark that on the other families the deduction requires some steps of the multiple premise rule F n and in this case both timings and growing ratio of EPDL-new are lower than EPDL. 6 Conclusions and future work The main novelty of the calculus presented in this paper is the presence of rules tailored to draw deductions about the (immediate) future and rules exploiting the information about facts known in the future to draw facts about the present state of knowledge. The aim is to reduce the branching of the proof to decide a formula in Dummett logic. In the presentation the focus is on rules to treat F -formulas, but rules for T formulas can be devised. As an example,, T(A B), F u A and, T(A B), F ub are rules reducing the number, T n B, F u A, FA, F u B of branches of the proofs, in particular when A is implicative. The next step of our investigation is to devise a calculus for intuitionistic logic having the same

15 signs. In particular, for the propositional fragment we are working on a calculus respecting the subformula property and that does not need any loop-checking mechanism. References 1. A. Avellone, M. Ferrari, and P. Miglioli. Duplication-free tableau calculi and related cut-free sequent calculi for the interpolable propositional intermediate logics. Logic Journal of the IGPL, 7(4): , A. Avellone, G. Fiorino, and U. Moscato. Optimization techniques for propositional intuitionistic logic and their implementation. Theoretical Computer cience, 409(1):41 58, A. Avron. imple consequence relations. Journal of Information and Computation, 92: , Arnon Avron and Beata Konikowska. Decomposition proof systems for gödeldummett logics. tudia Logica, 69(2): , M. Dummett. A propositional calculus with a denumerable matrix. Journal of ymbolic Logic, 24:96 107, G. Fiorino. An O(n log n)-space decision procedure for the propositional Dummett Logic. Journal of Automated Reasoning, 27(3): , G. Fiorino. Fast decision procedure for propositional dummett logic based on a multiple premise tableau calculus. Technical Report 164, Dipartimento di Metodi Quantitativi per le cienze Economiche ed Aziendali, Università degli tudi di Milano-Bicocca, M.C. Fitting. Intuitionistic Logic, Model Theory and Forcing. North-Holland, K. Gödel. On the intuitionistic propositional calculus. In. Feferman et al, editor, Collected Works, volume 1. Oxford University Press, Reiner Hähnle. Tableaux and related methods. In John Alan Robinson and Andrei Voronkov, editors, Handbook of Automated Reasoning, pages Elsevier and MIT Press, P. Hajek. Metamathematics of Fuzzy Logic. Kluwer, Dominique Larchey-Wendling. Combining proof-search and counter-model construction for deciding Gödel-Dummett logic. In Andrei Voronkov, editor, CADE, volume 2392 of Lecture Notes in Computer cience, pages pringer, Dominique Larchey-Wendling. Counter-model search in Gödel-Dummett logics. In David A. Basin and Michaël Rusinowitch, editors, IJCAR, volume 3097 of Lecture Notes in Computer cience, pages pringer, Dominique Larchey-Wendling. Graph-based decision for Gödel-Dummett logics. J. Autom. Reasoning, 38(1-3): , Fabio Massacci. implification: A general constraint propagation technique for propositional and modal tableaux. In Harrie de wart, editor, Proc. International Conference on Automated Reasoning with Analytic Tableaux and Related Methods, Oosterwijk, The Netherlands, volume 1397 of LNC, pages pringer- Verlag, R.M. mullyan. First-Order Logic. pringer, Berlin, 1968.