A Proof-Theoretic Semantic Analysis of Dynamic Epistemic Logic
Sabine Frittella, Giuseppe Greco, Alexander Kurz, Alessandra Palmigiano, Vlasta Sikimić
aa r X i v : . [ m a t h . L O ] M a y A Proof-Theoretic Semantic Analysisof Dynamic Epistemic Logic
Sabine Frittella ∗ Giuseppe Greco † Alexander Kurz ‡ Alessandra Palmigiano § Vlasta Sikimić ¶ Abstract
The present paper provides an analysis of the existing proof systems fordynamic epistemic logic from the viewpoint of proof-theoretic semantics.Dynamic epistemic logic is one of the best known members of a family oflogical systems which have been successfully applied to diverse scientificdisciplines, but the proof theoretic treatment of which presents many dif-ficulties. After an illustration of the proof-theoretic semantic principlesmost relevant to the treatment of logical connectives, we turn to illus-trating the main features of display calculi, a proof-theoretic paradigmwhich has been successfully employed to give a proof-theoretic semanticaccount of modal and substructural logics. Then, we review some of themost significant proposals of proof systems for dynamic epistemic logics,and we critically reflect on them in the light of the previously introducedproof-theoretic semantic principles. The contributions of the present pa-per include a generalisation of Belnap’s cut elimination metatheorem fordisplay calculi, and a revised version of the display-style calculus D.EAK[30]. We verify that the revised version satisfies the previously mentionedproof-theoretic semantic principles, and show that it enjoys cut elimina-tion as a consequence of the generalised metatheorem.
Keywords: display calculus, dynamic epistemic logic, proof-theoretic se-mantics.
Math. Subject Class. 2010: ∗ Laboratoire d’Informatique Fondamentale de Marseille (LIF) - Aix-Marseille Université. † Department of Values, Technology and Innovation - TU Delft. ‡ Department of Computer Science - University of Leicester. § Department of Values, Technology and Innovation - TU Delft. The research of the secondand fourth author has been made possible by the NWO Vidi grant 016.138.314, by the NWOAspasia grant 015.008.054, and by a Delft Technology Fellowship awarded in 2013 to thefourth author. ¶ The Institute of Philosophy of the Faculty of Philosophy - University of Belgrade. ontents A Special rules 45
A.1 Derived rules in D’.EAK . . . . . . . . . . . . . . . . . . . . . . . 45A.2 Soundness of comp rules in the final coalgebra . . . . . . . . . . . 47
B Cut elimination for D’.EAK 48C Completeness of D’.EAK 51 Introduction
In recent years, driven by applications in areas spanning from program semanticsto game theory, the logical formalisms pertaining to the family of dynamic logics [31, 48] have been very intensely investigated, giving rise to a proliferation ofvariants.Typically, the language of a given dynamic logic is an expansion of classicalpropositional logic with an array of modal-type dynamic operators , each of whichtakes an action as a parameter. The set of actions plays in some cases the roleof a set of indexes or parameters; in other cases, actions form a quantale-typealgebra. When interpreted in relational models, the formulas of a dynamic logicexpress properties of the model encoding the present state of affairs, as well asthe pre- and post-conditions of a given action. Actions formalize transformationsof one model into another one, the updated model, which encodes the state ofaffairs after the action has taken place.Dynamic logics have been investigated mostly w.r.t. their semantics andcomplexity, while their proof-theoretic aspects have been comparatively notso prominent. However, the existing proposals of proof systems for dynamiclogics witness a varied enough array of methodologies, that a methodologicalevaluation is now timely.The present paper is aimed at evaluating the current proposals of proof-systems for the best-known dynamic epistemic logics from the viewpoint of proof-theoretic semantics .Proof-theoretic semantics [47] is a theory of meaning which assigns formalproofs or derivations an autonomous semantic content. That is, formal proofsare treated as entities in terms of which meaning can be accounted for. Proof-theoretic semantics has been very influential in an area of research in structuralproof theory which aims at defining the meaning of logical connectives in termsof an analysis of the behaviour of the logical connectives inside the derivationsof a given proof system. Such an analysis is possible only in the context ofproof systems which perform well w.r.t. certain criteria; hence, one of the mainthemes in this area is to identify design criteria which both guarantee thatthe proof system enjoys certain desirable properties such as normalization orcut-elimination, and which make it possible to speak about the proof-theoreticmeaning for given logical connectives.An analysis of dynamic logics from a proof-theoretic semantic viewpoint isbeneficial both for dynamic logics and for structural proof theory. Indeed, suchan analysis provides dynamic logics with sound methodological and foundationalprinciples, and with an entirely novel perspective on the topic of dynamics andchange, which is independent from the dominating model-theoretic methods.Moreover, such an analysis provides structural proof theory with a novel ar-ray of case studies against which to test the generality of its proof-theoreticsemantic principles, and with the opportunity to extend its modus operandi tostill uncharted settings, such as the multi-type calculi introduced in [23].The structure of the paper goes as follows: in section 2, we introduce the basicideas of proof-theoretic semantics, as well as some of the principles in struc-3ural proof theory that were inspired by it, and we explain their consequencesand spirit, in view of their applications in the following sections. In section 3,we prove a generalisation of Belnap’s cut elimination metatheorem. In section4, we review some of the most significant proposals of proof systems for dy-namic epistemic logics, focusing mainly on the logic of Public Announcements(PAL) [44] and the logic of Epistemic Knowledge and Actions (EAK) [12], andwe critically reflect on them in the light of the principles of proof-theoretic se-mantics stated in section 2; in particular, in subsection 4.4, we focus on thedisplay-type calculus D.EAK for PAL/EAK introduced in [30]: we highlight itscritical issues—the main of which being that a smooth (
Belnap-style ) proof ofcut-elimination is not readily available for it. In section 5, we expand on thefinal coalgebra semantics for D.EAK, which will be relevant for the followingdevelopments. In section 6, we propose a revised version of D.EAK, discusswhy the revision is more adequate for proof-theoretic semantics, and finallyprove the cut-elimination theorem for the revised version as a consequence ofthe metatheorem proven in section 3. In section 7, we collect some conclusionsand indicate further directions. Most of the proofs and derivations are collectedin appendices A, B and C.
In the present section, we review and discuss the proof-theoretic notions whichwill be used in the further development of the paper. In the following subsection,we outline the conceptual foundations of proof-theoretic semantics; in subsection2.2, Belnap-style display calculi will be discussed; in subsection 2.3 a refinementof Belnap’s analysis, due to Wansing, will be reported on. Our presentation iscertainly not exhaustive, and will limit itself to targeting the issues needed inthe further development of the paper. The reader is referred to [47, 46] for adetailed presentation of proof-theoretic semantics, and to [49, 50] for a discussionof proof-theoretic semantic principles in structural proof theory.
Proof-theoretic semantics is a line of research which covers both philosophicaland technical aspects, and is concerned with methodological issues. Proof-theoretic semantics is based on the idea that a purely inferential theory ofmeaning is possible. That is, that the meaning of expressions (in a formallanguage or in natural language) can be captured purely in terms of the proofsand the inference rules which participate in the generation of the given expres-sion, or in which the given expression participates. This inferential view isopposed to the mainstream denotational view on the theory of meaning, andis influential in e.g. linguistics, linking up to the idea, commonly attributed toWittgenstein, that ‘meaning is use’. In proof theory, this idea links up with4entzen’s famous observation about the introduction and elimination rules ofhis natural deduction calculi:‘The introductions represent, as it were, the definitions of the sym-bols concerned, and the eliminations are no more, in the final ana-lysis, than the consequences of these definitions. This fact may beexpressed as follows: In eliminating a symbol, we may use the for-mula with whose terminal symbol we are dealing only in the senseafforded it by the introduction of that symbol’. ([25] p. 80)In the proof-theoretic semantic literature, this observation is brought to itsconsequences: rather than viewing proofs as entities the meaning of which isdependent on denotation, proof-theoretic semantics assigns proofs (in the senseof formal deductions) an autonomous semantic role ; that is, proofs are entitiesin terms of which meaning can be accounted for.Proof-theoretic semantics has inspired and unified much of the research instructural proof theory focusing on the purely inferential characterization oflogical constants (i.e. logical connectives) in the setting of a given proof system.
Display calculi are among the approaches in structural proof theory aimed atthe uniform development of an inferential theory of meaning of logical constantsaligned with the ideas of proof-theoretic semantics. Display calculi have beensuccessful in giving adequate proof-theoretic accounts of logics—such as modallogics and substructural logics—which have notoriously been difficult to treatwith other approaches. In particular, the contributions in this line of researchwhich are most relevant to our analysis are Belnap’s [15], Wansing’s [49], Goré’s[28], and Restall’s [45].
Display Logic.
Nuel Belnap introduced the first display calculus, which hecalls
Display Logic [15], as a sequent system augmenting and refining Gentzen’sbasic observations on structural rules. Belnap’s refinement is based on the in-troduction of a special syntax for the constituents of each sequent. Indeed, hiscalculus treats sequents X ⊢ Y where X and Y are so-called structures , i.e.syntactic objects inductively defined from formulas using an array of specialconnectives. Belnap’s basic idea is that, in the standard Gentzen formulation,the comma symbol ‘ , ’ separating formulas in the precedent and in the succedentof sequents can be recognized as a metalinguistic connective, of which the struc-tural rules define the behaviour.Belnap took this idea further by admitting not only the comma, but also sev-eral other connectives to keep formulas together in a structure, and called them structural connectives . Just like the comma in standard Gentzen sequents is in-terpreted contextually (that is, as conjunction when occurring on the left-handside and as disjunction when occurring on the right-hand side), each structural5onnective typically corresponds to a pair of logical connectives, and is inter-preted as one or the other of them contextually (more of this in sections 5and 6.1). Structural connectives maintain relations with one another, the mostfundamental of which take the form of adjunctions and residuations. These rela-tions make it possible for the calculus to enjoy the powerful property which givesit its name, namely, the display property . Before introducing it formally, let usagree on some auxiliary definitions and nomenclature: structures are definedmuch in the same way as formulas, taking formulas as atomic components andclosing under the given structural connectives; therefore, each structure can beuniquely associated with a generation tree. Every node of such a generation treedefines a substructure . A sequent X ⊢ Y is a pair of structures X, Y . The dis-play property was introduced by Belnap, see Theorem 3.2 of [15] (where X ⊢ Y is called a consecution and X the antecedent and Y the consequent): Definition 1.
A proof system enjoys the display property iff for every sequent X ⊢ Y and every substructure Z of either X or Y , the sequent X ⊢ Y can beequivalently transformed, using the rules of the system, into a sequent whichis either of the form Z ⊢ W or of the form W ⊢ Z , for some structure W . Inthe first case, Z is displayed in precedent position , and in the second case, Z is displayed in succedent position . The rules enabling this equivalent rewriting arecalled display postulates .Thanks to the fact that display postulates are based on adjunction andresiduation, in display calculi exactly one of the two alternatives mentioned inthe definition above occurs. In other words, in a system enjoying the displayproperty, any substructure of any sequent X ⊢ Y is always displayed either onlyin precedent position or only in succedent position. This is why we can talkabout occurrences of substructures in precedent or in succedent position, evenif they are nested deep within a given sequent, as illustrated in the followingexample: Y ⊢ X > ZX ; Y ⊢ ZY ; X ⊢ ZX ⊢ Y > Z
In the derivation above, the structure X is on the right side of the turnstile,but it is displayable on the left, and therefore is in precedent position. As wewill see next, the display property is a crucial technical ingredient for displaycalculi, but it is also at the basis of Belnap’s methodology for characterizingoperational connectives: according to Belnap, any logical connective should beintroduced in isolation , i.e., when it is introduced, the context on the side ithas been introduced must be empty. The display property guarantees that thiscondition is not too restrictive.To illustrate the fundamental role played by the display property in thetransformation steps of the cut elimination metatheorem, consider the elimina-6ion step of the following cut application, in which the cut formula is principalon both premises of the cut. ... π X ⊢ A ... π Y ⊢ BX ; Y ⊢ A ∧ B ... πA ; B ⊢ ZA ∧ B ⊢ ZX ; Y ⊢ Z ... π Y ⊢ B ... π X ⊢ A ... πA ; B ⊢ ZA ⊢ Z < BX ⊢ Z < BX ; B ⊢ ZB ⊢ X > ZY ⊢ X > ZX ; Y ⊢ Z The dashed lines in the prooftree on the right-hand side correspond to ap-plications of display postulates. Clearly, this transformation step has been madepossible because the display postulates disassemble, as it were, compound struc-tures so as to give us access to the immediate subformulas of the original cutformula, and then reassemble them so as to ‘put things back again’. Hence,it is possible to break down the original cut into two cut applications on theimmediate subformulas, as required by the original Gentzen strategy.
Canonical cut elimination.
In [15], a meta-theorem is proven, which givessufficient conditions in order for a sequent calculus to enjoy cut-elimination. This meta-theorem captures the essentials of the Gentzen-style cut-eliminationprocedure, and is the main technical motivation for the design of Display Lo-gic. Belnap’s meta-theorem gives a set of eight conditions on sequent calculi,which are relatively easy to check, since most of them are verified by inspectionon the shape of the rules. Together, these conditions guarantee that the cut iseliminable in the given sequent calculus, and that the calculus enjoys the subfor-mula property. When Belnap’s metatheorem can be applied, it provides a muchsmoother and more modular route to cut-elimination than the Gentzen-styleproofs. Moreover, as we will see later, a Belnap style cut-elimination theoremis robust with respect to adding structural rules and with respect to addingnew logical connectives, whereas a Gentzen-style cut-elimination proof for themodified system cannot be deduced from the old one, but must be proved fromscratch.In a slogan, we could say that Belnap-style cut-elimination is to ordinarycut-elimination what canonicity is to completeness: indeed, canonicity providesa uniform strategy to achieve completeness. In the same way, the conditionsrequired by Belnap’s meta-theorem ensure that one and the same given set oftransformation steps is enough to achieve Gentzen-style cut elimination for anysystem satisfying them. In what follows, we review and discuss eight conditions which are strongerin certain respects than those in [15], and which define the notion of proper Note that, as Belnap observed on pag. 389 in [15]: ‘The eight conditions are supposed tobe a reminiscent of those of Curry’ in [18]. The relationship between canonicity and Belnap-style cut-elimination is in fact more thana mere analogy, see [32, Theorem 20]. See also [16, 45] and the ‘second formulation’ of condition C6/7 in subsection 4.4 of [49]. isplay calculus in [49]. C : Preservation of formulas. This condition requires each formula oc-curring in a premise of a given inference to be the subformula of some formulain the conclusion of that inference. That is, structures may disappear, but notformulas. This condition is not included in the list of sufficient conditions ofthe cut-elimination meta-theorem, but, in the presence of cut-elimination, itguarantees the subformula property of a system. Condition C can be verifiedby inspection on the shape of the rules. C : Shape-alikeness of parameters. This condition is based on the re-lation of congruence between parameters (i.e., non-active parts) in inferences;the congruence relation is an equivalence relation which is meant to identify thedifferent occurrences of the same formula or substructure along the branchesof a derivation [15, section 4], [45, Definition 6.5]. Condition C requires thatcongruent parameters be occurrences of the same structure. This can be under-stood as a condition on the design of the rules of the system if the congruencerelation is understood as part of the specification of each given rule; that is,each rule of the system comes with an explicit specification of which elementsare congruent to which (and then the congruence relation is defined as the re-flexive and transitive closure of the resulting relation). In this respect, C isnothing but a sanity check, requiring that the congruence is defined in such away that indeed identifies the occurrences which are intuitively “the same”. C : Non-proliferation of parameters. Like the previous one, also this con-dition is actually about the definition of the congruence relation on parameters.Condition C requires that, for every inference (i.e. rule application), each ofits parameters is congruent to at most one parameter in the conclusion of thatinference. Hence, the condition stipulates that for a rule such as the following, X ⊢ YX ; X ⊢ Y the structure X from the premise is congruent to only one occurrence of X in the conclusion sequent. Indeed, the introduced occurrence of X should beconsidered congruent only to itself. Moreover, given that the congruence isan equivalence relation, condition C implies that, within a given sequent, anysubstructure is congruent only to itself. Remark 1.
Conditions C and C make it possible to follow the history of aformula along the branches of any given derivation. In particular, C impliesthat the the history of any formula within a given derivation has the shape of atree, which we refer to as the history-tree of that formula in the given derivation.Notice, however, that the history-tree of a formula might have a different shapethan the portion of the underlying derivation corresponding to it; for instance, See the ‘first formulation’ of conditions C6, C7 in subsection 4.1 of [49]. A which is absent in the underlying branch of the derivationtree, given that Contraction is a unary rule. ...A ; A ⊢ XA ⊢ X t t t t ❅❅❅(cid:0)(cid:0)(cid:0) t C : Position-alikeness of parameters. This condition bans any rule inwhich a (sub)structure in precedent (resp. succedent) position in a premise iscongruent to a (sub)structure in succedent (resp. precedent) position in theconclusion. C : Display of principal constituents. This condition requires that anyprincipal occurrence be always either the entire antecedent or the entire con-sequent part of the sequent in which it occurs. In the following section, ageneralization of this condition will be discussed, in view of its application tothe main focus of interest of the present paper.The following conditions C and C are not reported below as they are statedin the original paper [15], but as they appear in [49, subsection 4.1]. More aboutthis difference is discussed in section 7.2. C : Closure under substitution for succedent parameters. This condi-tion requires each rule to be closed under simultaneous substitution of arbitrarystructures for congruent formulas which occur in succedent position. ConditionC ensures, for instance, that if the following inference is an application of therule R : ( X ⊢ Y ) (cid:0) [ A ] suci | i ∈ I (cid:1) R ( X ′ ⊢ Y ′ )[ A ] suc and (cid:0) [ A ] suci | i ∈ I (cid:1) represents all and only the occurrences of A in the premisswhich are congruent to the occurrence of A in the conclusion , then also thefollowing inference is an application of the same rule R : ( X ⊢ Y ) (cid:0) [ Z/A ] suci | i ∈ I (cid:1) R ( X ′ ⊢ Y ′ )[ Z/A ] suc where the structure Z is substituted for A .This condition caters for the step in the cut elimination procedure in which thecut needs to be “pushed up” over rules in which the cut-formula in succedentposition is parametric. Indeed, condition C guarantees that, in the picturebelow, a well-formed subtree π [ Y /A ] can be obtained from π by replacing any Clearly, if I = ∅ , then the occurrence of A in the conclusion is congruent to itself. A corresponding to a node in the history tree of the cut-formula A by Y , and hence the following transformation step is guaranteed go throughuniformly and “canonically”: ... π ′ X ′ ⊢ A... π X ⊢ A ... π A ⊢ YX ⊢ Y ... π ′ X ′ ⊢ A ... π A ⊢ YX ′ ⊢ Y... π [ Y/A ] X ⊢ Y if each rule in π verifies condition C . C : Closure under substitution for precedent parameters. This condi-tion requires each rule to be closed under simultaneous substitution of arbitrarystructures for congruent formulas which occur in precedent position. ConditionC can be understood analogously to C , relative to formulas in precedent pos-ition. Therefore, for instance, if the following inference is an application of therule R : ( X ⊢ Y ) (cid:0) [ A ] prei | i ∈ I (cid:1) R ( X ′ ⊢ Y ′ )[ A ] pre then also the following inference is an instance of R : ( X ⊢ Y ) (cid:0) [ Z/A ] prei | i ∈ I (cid:1) R ( X ′ ⊢ Y ′ )[ Z/A ] pre Similarly to what has been discussed for condition C , condition C caters forthe step in the cut elimination procedure in which the cut needs to be “pushedup” over rules in which the cut-formula in precedent position is parametric. C : Eliminability of matching principal constituents. This conditionrequests a standard Gentzen-style checking, which is now limited to the casein which both cut formulas are principal , i.e. each of them has been introducedwith the last rule application of each corresponding subdeduction. In this case,analogously to the proof Gentzen-style, condition C requires being able totransform the given deduction into a deduction with the same conclusion inwhich either the cut is eliminated altogether, or is transformed in one or moreapplications of cut involving proper subformulas of the original cut-formulas. Rules introducing logical connectives.
In display calculi, these rules, some-times referred to as operational rules as opposed to the structural rules, typicallyoccur in two flavors: operational rules which translate one structural connective10n the premises in the corresponding connective in the conclusion, and opera-tional rules in which both the operational connective and its structural counter-part are introduced in the conclusion. An example of this pattern is providedbelow for the case of the modal operator ‘diamond’: ◦ A ⊢ X L A ⊢ X X ⊢ A R ◦ X ⊢ A This introduction pattern obeys very strict criteria, which will be expanded onin the next subsection. From this example, it is clear that the introduction rulescapture the rock bottom behavior of the logical connective in question; addi-tional properties (for instance, normality, in the case in point), which mightvary depending on the logical system, are to be captured at the level of ad-ditional (purely structural) rules. This enforces a clear-cut division of labourbetween operational rules, which only encode the basic proof-theoretic meaningof logical connectives, and structural rules, which account for all extra relationsand properties, and which can be modularly added or removed, thus accountingfor the space of logics.Summing up, the two main benefits of display calculi are a “canonical” proofof cut elimination, and an explicit and modular account of logical connectives.
In [49, subsubsection 1.3], referring to the well known idea that ‘a proof-theoreticsemantics exemplifies the Wittgensteinian slogan that meaning is use’, Wansingstresses that, for this slogan to serve as a conceptual basis for a general inferentialtheory of meaning, ‘use’ should be understood as ‘ correct use’. The consequencesof the idea of meaning as correct use then precipitate into the following principlesfor the introduction rules for operational connectives, which he discusses in thesame subsection and which are reported below. These principles are hence tobe understood as the general requirements a (sequent-style) proof system needsto satisfy in order to encode the correct use, and hence for being suitable forproof-theoretic semantics.
Separation.
This principle requires that the introduction rules for a givenconnective f should not exhibit any other connective rather than f . Hencethe meaning of a given operational connective cannot be dependent from anyother operational connectives. For instance, the following rule does not satisfy separation : Γ ⊢ A, ∆ Γ ⊢ A , ∆ This criterion does not ban the possibility of defining composite connectives;however, it ensures that the dependence relation between connectives createsno vicious circles. In fact, as it is formulated, this criterion is much stronger,since it requires every connective to be independent of any other.11 solation.
This is a stronger requirement than separation, and stipulates that,in addition, the precedent (resp. succedent) of the conclusion sequent in a left(resp. right) introduction rule must not exhibit any structure operation. In [15],Belnap explains this requirement by remarking that an introduction rule withnonempty context on the principal side would fail to account for the meaningof the logical connective involved in a context-independent way.
Segregation.
This is an even stronger requirement than isolation, and stipu-lates that, in addition, also the auxiliary formulas in the premise(s) must occurwithin an empty context. This property appears under the name of visibility in[14]. Weak symmetry.
This requirement stipulates that each introduction rule fora given connective f should either belong to a set of rules ( f ⊢ ) which introduce f on the left-hand side of the turnstile ⊢ in the conclusion sequent, or to a setof rules ( ⊢ f ) which introduce f on the right-hand side of the turnstile ⊢ in theconclusion sequent. Understanding the either-or as exclusive disjunction, thiscriterion prevents an operational connective to be introduced on both sides bythe application of one and the same rule. Thus, weak symmetry stipulates thatthe sets ( f ⊢ ) and ( ⊢ f ) be disjoint. However, weak symmetry does not excludethat either ( f ⊢ ) or ( ⊢ f ) be empty. Symmetry.
This condition strengthens weak symmetry by requiring both ( f ⊢ ) and ( ⊢ f ) to be nonempty for each connective f . Rather than a re-quirement on individual rules, this principle is a requirement on the set of theintroduction rules for any given connective. Notice that symmetry does notexclude the possibility of having, for instance, two rules that introduce a givenconnective on the left and one that introduces it on the right side of the turnstile. Weak explicitness.
An introduction rule for f is weakly explicit if f occursonly in the conclusion of a rule and not in its premisses. Explicitness.
An introduction rule for f is explicit if it is weakly explicit andin addition to this, f appears only once in the conclusion of the rule.The following principles are of a more global nature, which involves the proofsystem as a whole: Unique characterization.
This principle requires each logical connective tobe uniquely characterized by its behaviour in the system, in the following sense.Let Λ be a logical system with a syntactic presentation S in which f occurs. Let S ∗ be the result of rewriting f everywhere in S as f ∗ , and let ΛΛ ∗ be the system In [24], following ideas from [14], the visibility property has been identified as an essentialingredient to generalise Belnap’s metatheorem beyond display calculi. SS ∗ of S and S ∗ in the combined language with both f and f ∗ . Let A f denote a formula (in this language) that contains a certainoccurrence of f , and let A f ∗ denote the result of replacing this occurrence of f in A f by f ∗ . The connectives f and f ∗ are uniquely characterized in ΛΛ ∗ (cfr.[49, subsubsection 1.4]) if for every formula A f in the language of ΛΛ ∗ , A f isprovable in SS ∗ iff A f ∗ is provable in SS ∗ . Došen’s principle.
Hilbert style presentations are modular in the followingsense: if Λ and Λ are finitely axiomatizable logics over the same languageand Λ is stronger than Λ , then an axiomatization of Λ can be obtained fromone of Λ by adding finitely many axioms to it. This makes it possible tomodularly generate all finite axiomatic extensions of a given logic. Although itis arguably more difficult to achieve an analogous degree of modularity in thesequent calculi presentation, a principle aimed to achieve it has been advocatedby Wansing under the name of Došen’s principle (cfr. [49, subsubsection 1.5]):“The rules for the logical operations are never changed; all changes are made inthe structural rules”. Thus, suitable finite axiomatic extensions of a given logic L can be captured by adding structural rules to the proof system associated with L . Display calculi are particularly suitable to implement Došen’s principle. Asremarked early on, besides featuring structural rules which encode properties ofsingle structural connectives (which is the case e.g. of the rule exchange), displaycalculi typically feature rules which concern the interaction between differentstructural connectives (the adjunction between two structural connectives is anexample of the latter type of rule, see for instance the rules applied in theexample on page 6). Cut-eliminability.
Finally, Wansing considers the eliminability of the cut rule as an important requirement for the proof-theoretic semantics of logicalconnectives.
In the present section, we discuss a slight extension of Wansing’s notion ofproper display calculus (cf. Subsection 2.2), and prove its associated Belnap-style cut elimination metatheorem. The cut elimination for the calculus D’.EAKintroduced in Section 6.3 (see also Appendix B) will be derived as an instanceof the metatheorem below.
Definition 2.
A sequent calculus is a quasi proper display calculus if it verifiesconditions C , C , C , C , C , C , C of section 2.2, and moreover it satisfiesthe following conditions C ′ , C ′′ and C ′ :13 ′ : Quasi-display of principal constituents. If a formula A is principalin the conclusion sequent s of a derivation π , then A is in display, unless π consists only of its conclusion sequent s (i.e. s is an axiom). C ′′ : Display-invariance of axioms. If a display rule can be applied to anaxiom s , the result of that rule application is again an axiom. C ′ : Closure of axioms under cut. If X ⊢ A and A ⊢ Y are axioms, then X ⊢ Y is again an axiom.Notice that condition C in Subsection 2.2 is stronger than both C ′ and C ′′ ,and that the strength of condition C ′ is intermediate between that of C andof the following one, appearing in [45, Definition 6.8]: C ′′′ : Single principal constituents. This condition requires that, in theconclusion of any rule, there be at most one non-parametric formula—which isthe formula introduced by the application of the rule in question—unless therule is an axiom.The above condition C ′′′ is introduced in [45] within a setting accountingfor sequent calculi which do not necessarily enjoy the full display property. Thecalculi considered in [45] are such that the introduction rules do not need toenjoy the requirement of isolation (cf. Chapter 6), and the (multiple) cut ruleapplies at any depth. The calculus introduced in Section 6.1 enjoys the fulldisplay property, therefore the following cut rule, in which both cut formulasoccur in isolation: X ⊢ A A ⊢ Y CutX ⊢ Y will be taken as primitive in it without loss of generality, as is standardly done indisplay calculi. However, the calculus in Section 6.1 fails to enjoy the propertyof isolation , which typically plays a role in the cut elimination metatheoremfor display calculi, and indeed appears in [49] as condition C . In the nextsubsection, we show that, even when the cut rule is the one above, requiring thecombination of C ′ and C ′′ suffices. The aim of the present subsection is to prove the following theorem:
Theorem 1.
Any calculus satisfying conditions C , C , C , C ′ , C ′′ , C , C ,C , and C ′ enjoys cut elimination. If C is also satisfied, then the calculusenjoys the subformula property. In [23], we give a metatheorem which is based on a different tradeoff: on the one hand,we will not require the full display property, but on the other we will require a condition closeto segregation. roof. This is a generalization of the proof in [51, Section 3.3, Appendix A].For the sake of conciseness, we will expand only on the parts of the proof whichdepart from that treatment.Our original derivation is ... π X ⊢ A ... π A ⊢ YX ⊢ Y Principal stage: both cut formulas are principal.
There are three sub-cases.If the end sequent X ⊢ Y is identical to the conclusion of π (resp. π ), thenwe can eliminate the cut simply replacing the derivation above with π (resp. π ).If the premises X ⊢ A and A ⊢ Y are axioms, then, by C ′ , the conclusion X ⊢ Y is an axiom, therefore the cut can be eliminated by simply replacing theoriginal derivation with X ⊢ Y .If one of the two premises of the cut in the original derivation is not anaxiom, then, by C , there is a proof of X ⊢ Y which uses the same premise(s)of the original derivation and which involves only cuts on proper subformulasof A . Parametric stage: at least one cut formula is parametric.
There aretwo subcases: either one cut formula is principal or they are both parametric.Consider the subcase in which one cut formula is principal. W.l.o.g. weassume that the cut-formula A is principal in the the left-premise X ⊢ A of thecut in the original proof (the other case is symmetric). As discussed in Remark1, conditions C and C make it possible to consider the history-tree of theright-hand-side cut formula A in π . The situation can be pictured as follows: ... π X ⊢ A ... π .i A i ⊢ Y i . . . ... π .j ( X j ⊢ Y j )[ A j ] pre ... ... π .k ( X k ⊢ Y k )[ A k ] pre . . .. . . ... . . . π A ⊢ YX ⊢ Y where, for i, j, k ∈ { , . . . , n } , the nodes A i ⊢ Y i , ( X j ⊢ Y j )[ A j ] pre , and ( X k ⊢ Y k )[ A k ] pre represent the three ways in which the leaves A i , A j and A k in the history-tree of A in π can be introduced, and which will be discussed below. The notation A A ) indicates that the given occurrence is principal (resp. parametric).Notice that condition C guarantees that all occurrences in the history of A arein precedent position in the underlying derivation tree.Let A l be introduced as a parameter (as represented in the picture abovein the conclusion of π .k for A l = A k ). Assume that ( X k ⊢ Y k )[ A k ] is theconclusion of an application inf of the rule Ru (for instance, in the calculus ofsection 6.1, this situation arises if A k has been introduced with an applicationof Weakening). Since A k is a leaf in the history-tree of A , we have that A k iscongruent only to itself in X k ⊢ Y k . Hence, C implies that it is possible tosubstitute X for A k by means of an application of the same rule Ru . That is, ( X k ⊢ Y k )[ A k ] can be replaced by ( X k ⊢ Y k )[ X/A k ] .Let A l be introduced as a principal formula. The corresponding subcase in[51] splits into two subsubcases: either A l is introduced in display or it is not.If A l is in display (as represented in the picture above in the conclusion of π .i for A l = A i ), then we form a subderivation using π and π .i and applyingcut as the last rule.If A l is not in display (as represented in the picture above in the conclusionof π .j for A l = A j ), then condition C ′ implies that ( X j ⊢ Y j )[ A j ] pre is anaxiom (so, in particular, there is at least another occurrence of A in succedentposition), and C ′′ implies that some axiom A j ⊢ Y ′ j exists, which is display-equivalent to the first axiom, and in which A j occurs in display. Let π ′ bethe derivation which transforms A j ⊢ Y ′ i into ( X j ⊢ Y j )[ A j ] pre . We form asubderivation using π and A j ⊢ Y ′ j and joining them with a cut application,then attaching π ′ [ X/A j ] pre below the new cut.The transformations just discussed explain how to transform the leaves ofthe history tree of A . Finally, condition C implies that substituting X for eachoccurrence of A in the history tree of the cut formula A in π (or in a display-equivalent proof π ′ ) gives rise to an admissible derivation π [ X/A ] pre (use C for the symmetric case).Summing up, this procedure generates the following proof tree: ... π X ⊢ A ... π .i A i ⊢ Y i X ⊢ Y i . . . ... π X ⊢ A A j ⊢ Y ′ [ A ] suc X ⊢ Y ′ [ A ] suc ... π ′ [ X/A ] pre ( X j ⊢ Y j )[ X/A j ] pre [ A ] suc ... ... π .k ( X k ⊢ Y k )[ X/A k ] pre . . .. . . ... . . . π [ X/A ] pre X ⊢ Y A (in theright-hand-side premise of the given cut application) contains at most one leaf A l which is principal, then the height of the new cuts is lower than the heightof the original cut.If, in the original derivation, the history-tree of the cut formula A (in theright-hand-side premise of the given cut application) contains more than one leaf A l which is principal, then we cannot conclude that the height of the new cutsis always lower than the height of the original cut (for instance, in the calculusintroduced in Section 6.1, this situation may arise when two ancestors of a cutformula are introduced as principal, and then are identified via an applicationof the rule Contraction). In this case, we observe that in each newly introducedapplication of the cut rule, both cut formulas are principal. Hence, we canapply the procedure described in the Principal stage and transform the originalderivation in a derivation in which the cut formulas of the newly introducedcuts have strictly lower complexity than the original cut formula.Finally, as to the subcase in which both cut formulas are parametric, considera proof with at least one cut. The procedure is analogous to the previous case.Namely, following the history of one of the cut formulas up to the leaves, andapplying the transformation steps described above, we arrive at a situation inwhich, whenever new applications of cuts are generated, in each such applicationat least one of the cut formulas is principal. To each such cut, we can apply(the symmetric version of) the Parametric stage described so far. In the present section, we first review the two best known logical systems in thefamily of dynamic epistemic logics, namely public announcement logic (PAL)[44], and the logic of epistemic actions and knowledge (EAK) [12], focusingmainly on the latter one. Our presentation in subsection 4.1 is different butequivalent to the original version from [12] (without common knowledge), andrather follows the presentation given in [39] and in [30]. In subsections 4.3and 4.4 we discuss their existing proof-theoretic formalizations, particularly inrelation to the viewpoint of proof-theoretic semantics, and mention the systemD.EAK as a promising approximation of a setting for proof-theoretic semantics.Finally, in subsection 5, we discuss the final coalgebra semantics, since this is asemantic environment in which all connectives of the language of D.EAK (andof its improved version D’.EAK) can be naturally interpreted.
The logic of epistemic actions and knowledge (further on EAK) is a logicalframework which combines a multi-modal classical logic with a dynamic-typepropositional logic. Static modalities in EAK are parametrized with agents,and their intended interpretation is epistemic, that is, h a i A intuitively stands17or ‘agent a thinks that A might be the case’. Dynamic modalities in EAKare parametrized with epistemic action-structures (defined below) and theirintended interpretation is analogous to that of dynamic modalities in e.g. Pro-positional Dynamic Logic. That is, h α i A intuitively stands for ‘the action α isexecutable, and after its execution A is the case’. Informally, action structuresloosely resemble Kripke models, and encode information about epistemic actionssuch as e.g. public announcements, private announcements to a group of agents,with or without (actual or suspected) wiretapping, etc. Action structures con-sist of a finite nonempty domain of action-states, a designated state, binaryrelations on the domain for each agent, and a precondition map. Each statein the domain of an action structure α represents the possible appearance ofthe epistemic action encoded by α . The designated state represents the actionactually taking place. Each binary relation of an action structure representsthe type, or degree, of uncertainty entertained by the agent associated with thegiven binary relation about the action taking place; for instance, the agents’knowledge, ignorance, suspicions. Finally, the precondition function maps eachstate in the domain to a formula, which is intended to describe the state ofaffairs under which it is possible to execute the (appearing) action encoded bythe given state. This formula encodes the preconditions of the action-state. Thereader is referred to [12] for further intuition and concrete examples.Let AtProp be a countable set of atomic propositions, and Ag be a nonemptyset (of agents). The set L of formulas A of the logic of epistemic actions andknowledge (EAK), and the set Act ( L ) of the action structures α over L aredefined simultaneously as follows: A := p ∈ AtProp | ¬ A | A ∨ A | h a i A | h α i A ( α ∈ Act ( L ) , a ∈ Ag ) , where an action structure over L is a tuple α = ( K, k, ( α a ) a ∈ Ag , P re α ) , suchthat K is a finite nonempty set, k ∈ K , α a ⊆ K × K and P re α : K → L .The symbol P re ( α ) stands for P re α ( k ) . For each action structure α andevery i ∈ K , let α i := ( K, i, ( α a ) a ∈ Ag , P re α ) . Intuitively, the family of actionstructures { α i | kα a i } encodes the uncertainty of agent a about the action α = α k that is actually taking place. Perhaps the best known epistemic actionsare public announcements , formalized as action structures α such that K = { k } ,and α a = { ( k, k ) } for all a ∈ Ag . The logic of public announcements (PAL) [44]can then be subsumed as the fragment of EAK restricted to action structuresof the form described above. The connectives ⊤ , ⊥ , ∧ , → and ↔ are defined asusual.Standard models for EAK are relational structures M = ( W, ( R a ) a ∈ Ag , V ) such that W is a nonempty set, R a ⊆ W × W for each a ∈ Ag , and V : AtProp →P ( W ) . The interpretation of the static fragment of the language is standard.For every Kripke frame F = ( W, ( R a ) a ∈ Ag ) and each action structure α , let theKripke frame ` α F := ( ` K W, (( R × α ) a ) a ∈ Ag ) be defined as follows: ` K W isthe | K | -fold coproduct of W (which is set-isomorphic to W × K ), and ( R × α ) a is a binary relation on ` K W defined as ( w, i )( R × α ) a ( u, j ) iff wR a u and iα a j. M and each action structure α , let a α M := ( a α F , a K V ) be such that ` α F is defined as above, and ( ` K V )( p ) := ` K V ( p ) for every p ∈ AtProp . Finally, let the update of M with the action structure α be thesubmodel M α := ( W α , ( R α a ) a ∈ Ag , V α ) of ` α M the domain of which is thesubset W α := { ( w, j ) ∈ a K W | M, w (cid:13)
P re α ( j ) } . Given this preliminary definition, formulas of the form h α i A are interpreted asfollows: M, w (cid:13) h α i A iff M, w (cid:13)
P re α ( k ) and M α , ( w, k ) (cid:13) A. The model M α is intended to encode the (factual and epistemic) state ofaffairs after the execution of the action α . Summing up, the construction of M α is done in two stages: in the first stage, as many copies of the original model M are taken as there are ‘epistemic potential appearances’ of the given action(encoded by the action states in the domain of α ); in the second stage, statesin the copies are removed if their associated original state does not satisfy thepreconditions of their paired action-state.A complete axiomatization of EAK consists of copies of the axioms and rulesof the minimal normal modal logic K for each modal operator, either epistemicor dynamic, plus the following (interaction) axioms: h α i p ↔ ( P re ( α ) ∧ p ); (1) h α i¬ A ↔ ( P re ( α ) ∧ ¬h α i A ); (2) h α i ( A ∨ B ) ↔ ( h α i A ∨ h α i B ); (3) h α ih a i A ↔ ( P re ( α ) ∧ _ {h a ih α i i A | kα a i } ) . (4)The interaction axioms above can be understood as attempts at defining themeaning of any given dynamic modality h α i in terms of its interaction with theother connectives. In particular, while axioms (2) and (3) occur also in otherdynamic logics such as PDL, axioms (1) and (4) capture the specific behaviour ofepistemic actions. Specifically, axiom (1) encodes the fact that epistemic actionsdo not change the factual state of affairs, and axiom (4) plausibly rephrases thefact that ‘after the execution of α , agent a thinks that A might be the case’ interms of ‘there being some epistemic appearance of α to a such that a thinksthat, after its execution, A is the case’. An interesting aspect of these axiomsis that they work as rewriting rules which can be iteratively used to transformany EAK-formula into an equivalent one free of dynamic modalities. Hence, thecompleteness of EAK follows from the completeness of its static fragment, andEAK is not more expressive than its static fragment. However, and interestingly,there is an exponential gap in succinctness between equivalent formulas in thetwo languages [38]. 19ction structures are one among many possible ways to represent actions.Following [30], we prefer to keep a black-box perspective on actions, and toidentify agents a with the indistinguishability relation they induce on actions;so, in the remainder of the article, the role of the action-structures α i for kαi will be played by actions β such that α a β , allowing us to reformulate (4) as h α ih a i A ↔ ( P re ( α ) ∧ _ {h a ih β i A | α a β } ) . In [39, 35], an analysis of PAL and EAK has been given from the point of view ofalgebraic semantics, resulting in the definition of the intuitionistic counterpartsof PAL and EAK. In the present subsection, we briefly review the definitionof the latter one, as it reveals a more subtle interaction between the variousmodalities, thus preparing the ground for the even richer picture that will arisefrom the proof-theoretic analysis.Let
AtProp be a countable set of atomic propositions, and let Ag be anonempty set (of agents). The set L (m-IK) of the formulas A of the multi-modal version m-IK of Fischer Servi’s intuitionistic modal logic IK are induct-ively defined as follows: A := p ∈ AtProp | ⊥ | A ∨ A | A ∧ A | A → A | h a i A | [ a ] A ( a ∈ Ag ) The logic m-IK is the smallest set of formulas in the language L (m-IK)(where ¬ A abbreviates as usual A → ⊥ ) containing the following axioms andclosed under modus ponens and necessitation rules: Axioms A → ( B → A )( A → ( B → C )) → (( A → B ) → ( A → C )) A → ( B → A ∧ B ) A ∧ B → AA ∧ B → BA → A ∨ BB → A ∨ B ( A → C ) → (( B → C ) → ( A ∨ B → C )) ⊥ → A [ a ]( A → B ) → ([ a ] A → [ a ] B ) h a i ( A ∨ B ) → h a i A ∨ h a i B ¬h a i⊥ FS1 h a i ( A → B ) → ([ a ] A → h a i B ) FS2 ( h a i A → [ a ] B ) → [ a ]( A → B ) Inference Rules
MP if ⊢ A → B and ⊢ A , then ⊢ B Nec if ⊢ A , then ⊢ [ a ] A
20o define the language of the intuitionistic counterpart of EAK, let
AtProp bea countable set of atomic propositions, and let Ag be a nonempty set. The set L (IEAK) of the formulas A of the intuitionistic logic of epistemic actions andknowledge (IEAK), and the set Act ( L ) of the action structures α over L aredefined simultaneously as follows: A := p ∈ AtProp | ⊥ | A → A | A ∨ A | A ∧ A | h a i A | [ a ] A | h α i A | [ α ] A, where a ∈ Ag , and an action structure α over L (IEAK) is defined in just thesame way as action structures in section 4.1. Then, the logic IEAK is defined ina Hilbert-style presentation which includes the axioms and rules of m-IK plusthe Fischer Servi axioms FS1 and FS2 for each dynamic modal operator, plusthe following axioms and rules: Interaction Axioms h α i p ↔ P re ( α ) ∧ p [ α ] p ↔ P re ( α ) → p h α i⊥ ↔ ⊥h α i⊤ ↔ P re ( α )[ α ] ⊤ ↔ ⊤ [ α ] ⊥ ↔ ¬ P re ( α )[ α ]( A ∧ B ) ↔ [ α ] A ∧ [ α ] B h α i ( A ∧ B ) ↔ h α i A ∧ h α i B h α i ( A ∨ B ) ↔ h α i A ∨ h α i B [ α ]( A ∨ B ) ↔ P re ( α ) → ( h α i A ∨ h α i B ) h α i ( A → B ) ↔ P re ( α ) ∧ ( h α i A → h α i B )[ α ]( A → B ) ↔ h α i A → h α i B h α ih a i A ↔ P re ( α ) ∧ W {h a ih β i A | α a β } [ α ] h a i A ↔ P re ( α ) → W {h a ih β i A | α a β } [ α ][ a ] A ↔ P re ( α ) → V { [ a ][ β ] A | α a β }h α i [ a ] A ↔ P re ( α ) ∧ V { [ a ][ β ] A | α a β } Inference Rules
Nec if ⊢ A , then ⊢ [ α ] A In the present subsection, we discuss the most relevant existing proof-theoreticaccounts [9, 42, 41, 11, 19, 6, 7, 8] for the logic of public announcements [44]21nd for the logic of epistemic actions and knowledge [12].
Labelled tableaux for PAL.
In [9], a labelled tableaux system is proposedfor public announcement logic. This system is sound and complete with re-spect to the semantics of PAL. Moreover, the computational complexity of thistableaux system is shown to be optimal for satisfiability checking in the lan-guage of PAL. The system manipulates triples, called labelled formulas, of theform h µ, n, φ i such that µ is a (possibly empty) list of PAL-formulas, n is anatural number, and φ is a PAL-formula. Intuitively, the tuple h µ, n i standsfor an epistemic state of the model updated with a sequence of announcementsencoded by µ . To give a closer impression of this tableaux system, consider thefollowing rule: h ( α , ..., α k ) , n, ¬ K a A i R b K n ′ fresh h ǫ, n ′ , ¬ [ α ] ... [ α k ] A i : h a, n, n ′ i This rule can be read as follows: if a state n does not satisfy K a A afterthe sequence of announcements α , ..., α k , then at least one of its R a -successorstates n ′ in the original model, represented by the tuple h ǫ, n ′ i in the rule, mustsurvive the updates and not satisfy A . Hence, h ǫ, n ′ i must satisfy the formula h α i ... h α k i¬ A , which is classically equivalent to ¬ [ α ] ... [ α k ] A. Clearly, rules such as this one incorporate the relational semantics of PAL.This is not satisfactory from the point of view of proof-theoretic semantics,since it prevents these rules from providing an independent contribution to themeaning of the logical connectives. A second issue, of a more technical nature,is that the statement of this rule is grounded on the classical interdefinabilitybetween the box-type and diamond-type modalities. This implies that if wedispense with the classical propositional base, we would need to reformulatethis rule. Hence the calculus is non-modular in the sense discussed in section2.3.
Labelled sequent calculi for PAL.
In [42] and [41], cut-free labelled se-quent calculi for PAL are introduced with truthful and non-truthful announce-ments, respectively. Also in this case, the statement of the rules of these calculiincorporates the relational semantics. For instance, this is illustrated here belowfor the case of truthful announcements. w : µ,α A, w : µ [ α ] A, w : µ α, Γ ⊢ ∆ L [ ]: µ w : µ [ α ] A, w : µ α, Γ ⊢ ∆ w : µ α, Γ ⊢ ∆ , w : µ,α A R [ ]: µ Γ ⊢ ∆ , w : µ [ α ] A In the rules above, symbols such as w : µ A can be rearranged and thenunderstood as the labelled formulas h µ, w, A i in the tableaux system presentedbefore. The only difference is that w is an individual variable which stands fora given state of a relational structure, and not for a natural number; however,this difference is completely nonessential. Under this interpretation, it is clear22hat e.g. the rule L [ ]: µ encodes the relational satisfaction clause of [ α ] A , when α is a truthful announcement. The following rules are also part of the calculi. v : A, w : K a A, wR a v, Γ ⊢ ∆ LK a w : K a A, wR a v, Γ ⊢ ∆ wR a v, Γ ⊢ ∆ , v : A RK a Γ ⊢ ∆ , w : K a A Besides the individual variables w and v , the rules above feature the binaryrelation symbol R a encoding the epistemic uncertainty of the agent a . Sincethe relational semantics is imported in the definitions of the rules, the sameissues pointed out in the case of the tableaux system appear also here. Onthe other hand, importing the relational semantics allows for some remarkableextra power. Indeed, the interaction axiom (4) can be derived from the four rulesabove, which deal with static and dynamic modalities in complete independenceof one another. Merging different logics.
In [11] and [19], sequent calculi have been definedfor dynamic logics arising in an algebraic way, motivated by program semantics,with a methodology introduced by [1]. Essentially, this approach is based onthe idea of merging a linear-type logic of actions (more precisely, [40]) with aclassical or intuitionistic logic of propositions. Following the treatment of [1],this logic arises semantically as the logic of certain quantale-modules, namelyof maps ⋆ : M × Q → M , preserving complete joins in each coordinate, where Q is a quantale and M is a complete join-semilattice. Each q ∈ Q inducesa completely join-preserving operation ( − ⋆ q ) : M → M , which, by generalorder-theoretic facts, has a unique right adjoint [ q ] : M → M . That is, for every m, m ′ ∈ M , m ⋆ q ≤ m ′ iff m ≤ [ q ] m ′ . (5)Intuitively, the elements of Q are actions (or rather, inverses of actions), and M is an algebra interpreting propositions, which in the best known cases arisesas the complex algebra of some relational structure, and therefore will be e.g. acomplete and atomic Boolean algebra with operators. Thus the framework of[11] and [19] is vastly more general than dynamic epistemic logic as it is usuallyunderstood. A remarkable feature of this setting is that the dynamic operationswhich are intended as the interpretation of the primitive dynamic connectivesarise in this setting as adjoints of “more primitive” operations; thus, and muchmore importantly, every dynamic modality comes with its adjoint. Moreover,every epistemic modality (parametrized as usual with an agent) comes in twocopies: one as an operation on Q and one as an operation on M , and thesetwo copies are stipulated to interact in a suitable way. More formally, thesemantic structures are defined as tuples ( M, Q, { f A } A ∈ Ag ) , where M and Q are as above, and for every agent A , f A is a pair of completely join preservingmaps ( f MA : M → M, f QA : Q → Q ) such that the following three conditionshold: f QA ( q · q ′ ) ≤ f QA ( q ) · f QA ( q ′ ) (6)23 MA ( m ⋆ q ) ≤ f MA ( m ) ⋆ f QA ( q ) (7) ≤ f QA (1) . (8)Intuitively, for every agent A , the operation f MA is the diamond-type modaloperator encoding the epistemic uncertainty of A , and f QA is the diamond-typemodal operator encoding the epistemic uncertainty of A about the action thatis actually taking place. Given this understanding, condition (7) hardcodes thefollowing well-known DEL-axiom in the semantic structures above: ^ { [ A ][ q ′ ] m | qAq ′ } ⊢ [ q ][ A ] m. (9)where the notation qAq ′ means that the action q ′ is indistinguishable from q for the agent A . In (7), the element f QA ( q ) encodes the join of all such actions.Because ⋆ is bilinear, we get: f MA ( m ) ⋆ f QA ( q ) = f MA ( m ) ⋆ _ Q { q ′ | qAq ′ } = _ M { f MA ( m ) ⋆ q ′ | qAq ′ } . Hence, (7) can be equivalently rewritten in the form of a rule as follows: W { f MA ( m ) ⋆ q ′ | qAq ′ } ⊢ m ′ f MA ( m ⋆ q ) ⊢ m ′ Applying adjunction to the premise and to the conclusion gets us to: m ⊢ V { [ A ][ q ′ ] m ′ | qAq ′ } m ⊢ [ q ][ A ] m ′ Finally, rewriting the rule above back as an inequality gets us to (9). The firstpioneering proposal is the sequent calculus developed in [11]. This calculusmanipulates two kinds of sequents: Q-sequents, of the form Γ ⊢ Q q , where q isan action and Γ is a sequence of actions and agents, and M-sequents, of the form Γ ⊢ M m , where m is a proposition and Γ is a sequence of propositions, actionsand agents. These different entailment relations need to be brought together bymeans of rules of hybrid type, such as the left one below. m ′ ⊢ M m Γ Q ⊢ Q q Dy L [ q ] m ′ , Γ Q ⊢ M m Γ , q ⊢ M m Dy R Γ ⊢ M [ q ] m As to the soundness of the rule
DyL , let us identify the logical symbolswith their interpretation, assume that the inequalities m ≤ m ′ and Γ Q ≤ q aresatisfied on given M and Q respectively, and prove that [ q ] m ′ , Γ Q ≤ m in M .Indeed, [ q ] m ′ ⋆ Γ Q ≤ [ q ] m ′ ⋆ q ≤ m ′ ≤ m. where Γ Q now stands for a suitable product in Q of the interpretations of its individualcomponents. Γ Q ≤ q and ⋆ being order-preserving in itssecond coordinate; the second inequality is obtained by applying the right-to-left direction of (5) to the inequality [ q ] m ′ ≤ [ q ] m ′ ; the last inequality holdsby assumption. The soundness of Dy R follows likewise from the left-to-rightdirection of (5).This calculus is shown to be both sound and complete w.r.t. this algebraic se-mantics. The setting illustrated above is powerful enough that sufficiently manyepistemic actions can be encoded in it to support the formalisation of variousvariants of the Muddy Children Puzzle in which children might be cheating.However, cut-elimination for this system has not been proven.In [19], a similar framework is presented which exploits the same basic ideas,and results in a system with more explicit proof-theoretic performances andwhich is shown to be cut-free. However, like its previous version, this systemfocuses on a logic semantically arising from an algebraic setting which is vastlymore general than the usual relational setting. The issue about how it preciselyrestricts to the usual setting, and hence how the usual DEL-type logics can becaptured within this more general calculus, is left largely implicit. The semanticsetting of [11], where propositions are interpreted as elements of a right module M on a quantale Q , specialises in [19] to a setting in which M = ( A , { A , (cid:7) A : A ∈ Ag } ) , where A is a Heyting algebra and, for every agent A , the modalities A and (cid:7) A are adjoint to each other. Notice that A , which in the classicalcase is defined as ¬ A ¬ , cannot be expressed any more in this way, and needsto be added as a primitive connective, which has not been done in [19].As mentioned before, the design of this calculus gives a more explicit ac-count than its previous version to certain technical aspects which come fromthe semantic setting; for instance, the semantic setting motivating both papersfeatures two domains of interpretation (one for the actions and one for the pro-positions), which are intended to give rise to two consequence relations whichare to be treated on a par and then made to interact. In [11], the calculus ma-nipulates sequents which are made of heterogeneous components. For instance,in action-sequents Γ ⊢ Q q , the precedent Γ is a sequence in which both actionsand agents may occur. Since Γ is to be semantically interpreted as an elementof Q , they need to resort to a rather clumsy technical solution which consistsin interpreting, e.g. the sequence ( q, A, q ′ ) as the element f QA ( q ) · q ′ . In [19],the calculus is given in a deep-inference format; namely, rules of this calculusmake it possible to manipulate formulas inside a given context. This more ex-plicit bookkeeping makes it possible to prove the cut-elimination, following theoriginal Gentzen strategy. However, the presence of two different consequencerelations and the need to account for their interaction calls for the developmentof an extensive theory-of-contexts, in which no less than five different types ofcontexts need to be introduced. This also causes a proliferation of rules, sincethe possibility of performing some inferences depends on the type of contextunder which they are to be performed.25 alculi for updates. In [6], a formal framework accounting for dynamicrevisions or updates is introduced, in which the revisions/updates are formal-ized using the turnstile symbol. This framework has aspects similar to Hoarelogic: indeed, it manipulates sequent-type structures of the form φ, φ ′ | = φ ′′ ,such that φ and φ ′′ are formulas of proposition-type, and φ ′ is a formula ofevent-type. This formalism has also common aspects to [11] and [19]: indeed,both proposition-type and event-type (i.e. action-type) formulas allow epistemicmodalities for each agent, respectively accounting for the agent’s epistemic un-certainty about the world and about the actions actually taking place.In [8] and [7], three formal calculi are introduced, manipulating the syn-tactic structures above. Given that the turnstile encodes the update ratherthan a consequence relation or entailment, the syntactic structures above arenot sequents in a proper sense. Rather than sequent calculi, these calculi shouldbe rather regarded as being of natural deduction-type. As such, the design ofthese calculi presents many issues from a proof-theoretic semantic viewpoint;to mention only one, multiple connectives are introduced at the same time, forinstance in the following rule: φ, φ ′ ⊢ φ ′′ R . h Bj i ( φ ∧ P re ( p ′ )) , h Bj i ( φ ′ ∧ p ′ ) ⊢ h Bj i φ ′′ These calculi are shown to be sound and complete w.r.t. three semanticconsequence relations, respectively.
In [30], a display-style sequent calculus D.EAK has been introduced, which issound with respect to the final coalgebra semantics (cf. section 5), and completew.r.t. EAK, of which it is a conservative extension. Moreover, Gentzen-style cutelimination holds for D.EAK. Finally, this system is defined independently ofthe relational semantics of EAK, and therefore is suitable for a fine-grainedproof-theoretic semantic analysis.Here below, we are not going to report on it in detail, but we limit ourselves tomention the structural rules which capture the specific features of EAK:
Structural Rules with Side Conditions
P re ( α ) ; { α } A ⊢ X reduce L { α } A ⊢ X X ⊢ P re ( α ) > { α } A reduce R X ⊢ { α } AP re ( α ) ; { α }{ a } X ⊢ Y swap-in L P re ( α ) ; { a }{ β } α a β X ⊢ Y Y ⊢ P re ( α ) > { α }{ a } X swap-in R Y ⊢ P re ( α ) > { a }{ β } α a β X (cid:16) P re ( α ) ; { a }{ β } X ⊢ Y | α a β (cid:17) swap-out L P re ( α ) ; { α }{ a } X ⊢ ; (cid:16) Y | α a β (cid:17) (cid:16) Y ⊢ P re ( α ) > { a }{ β } X | α a β (cid:17) swap-out R ; (cid:16) Y | α a β (cid:17) ⊢ P re ( α ) > { α }{ a } X swap-out rules do not have a fixed arity; they have as many premises as thereare actions β such that α a β . In the conclusion, the symbol ; (cid:16) Y | α a β (cid:17) refersto a string ( · · · ( Y ; Y ) ; · · · ; Y ) with n occurrences of Y , where n = |{ β | α a β }| . Operational Rules with Side Conditions
P re ( α ) ; { α } A ⊢ X reverse L P re ( α ) ; [ α ] A ⊢ X X ⊢ P re ( α ) > { α } A reverse R X ⊢ P re ( α ) > h α i A The main issues of D.EAK from the point of view of Wansing’s criteriaare linked with the presence of the formula
P re ( α ) : namely, the swap-in and swap-out rules violate the principle that all parametric variables should occurunrestricted. Indeed, the occurrences of the formula P re ( α ) in these rules iseasily seen to be parametric, since P re ( α ) occurs both in the premises and inthe conclusion. Since P re ( α ) is (the metalinguistic abbreviation of) a formula,it is a structure of a very restricted shape. As to the swap-out rules, it is notdifficult to see, e.g. semantically (cf. [35, Definition 4.2.]), that the occurrencesof P re ( α ) can be removed both in the premises and in the conclusion withoutaffecting either the soundness of the rule or the proof power of the system; thisentirely remedies the problem. Likewise, as to swap-in , it is not difficult tosee that the occurrences of P re ( α ) can be removed in the premises, but not inthe conclusion. However, even modified in this way, the swap-in rules wouldnot be satisfactory. Indeed, the new form of swap-in would introduce P re ( α ) in the conclusion. Since P re ( α ) is a metalinguistic abbreviation of a formulawhich as such has no other specific restrictions, the occurrence of P re ( α ) in theconclusion of swap-in must also be regarded as parametric. However, we stillwould not be able to substitute arbitrary structures for it, which is the source ofthe problem. This problem would be solved if P re ( α ) could be expressed, as astructure, purely in terms of the parameter α and structural constants (but nostructural variables). If this was the case, swap-in would encode the relationsbetween all these logical constants, and all the occurring structural variableswould be unrestricted.Secondly, the rules reduce violate condition C : indeed, in each of them, aformula in the premisses, namely P re ( α ) , is not a subformula of any formulaoccurring in the conclusion. Together with the cut-elimination, condition C guarantees the subformula property (cf. [15, Theorem 4.3]), but is not itselfessential for the cut-elimination, and indeed, cut-elimination has been provenfor D.EAK (albeit not à la Belnap). The specific way in which reduce violates C is also not a very serious one. Indeed, if the formula P re ( α ) could be expressedin a structural way, this violation would disappear.This solution cannot be implemented in D.EAK because the language ofD.EAK does not have enough expressivity to talk about P re ( α ) in any otherway than as an arbitrary formula, which needs to be introduced via weakeningor via identity (if atomic). Being able to account for P re ( α ) in a satisfactoryway from a proof-theoretic semantic perspective would require being able tostate rules which, for any α , would introduce P re ( α ) specifically , thus capturing27ts proof-theoretic meaning. Thus, by having structural and operational rulesfor P re ( α ) , we would solve many problems in one stroke: on the one hand,we would gain the practical advantage of achieving the satisfaction of C , thusguaranteeing the subformula property; on the other hand, and more importantly,from a methodological perspective, we would be able to have a setting in whichthe occurrences of P re ( α ) are not to be regarded as side formulas, but rather,they would occur as structures , on a par with all the other structures they wouldbe interacting with.Finally, the only operational rules violating Wansing’s separation principle(cf. subsection 2.3) are the reverse rules: P re ( α ); { α } A ⊢ Xrev L P re ( α ); [ α ] A ⊢ X X ⊢ P re ( α ) > { α } A rev R X ⊢ P re ( α ) > h α i A Here again, the problem comes from the fact that the language is not expressiveenough to capture the principles encoded in the rules above at a purely structurallevel. In this operational formulation, these rules are to participate, in our viewimproperly, in the proof-theoretic meaning of the connectives [ α ] and h α i . Thus,it would be desirable that the rules above could be either derived, so that theydisappear altogether, or alternatively, be reformulated as structural rules. In order to provide a justification for the soundness of the display postulatesinvolving the dynamic connectives, in [30] the final coalgebra was used as asemantic environment for the calculus D.EAK. Specifically, the final coalgebrawas there used to show that D.EAK is sound, and conservatively extends EAK.In the present section, we briefly review the needed preliminaries on the finalcoalgebra, and then the interpretation of EAK-formulas in the final coalgebra,which we will use in section 6.2 to show that D’.EAK is sound, and conservativelyextends EAK. The general notion of a coalgebra, as an arrow W → F W is given w.r.t. a functor F : C → C on an arbitrary category C , and much of thetheory of coalgebras is devoted to establishing results on coalgebras parametricin that functor F . For example, important notions such as bisimilarity andHennessy-Milner logics can be given for arbitrary functors on the category ofsets (and many other concrete categories). But even if one is interested, as in This semantics specifically applies to the classical base. Analogous ideas can be developedfor weaker propositional bases, but in the present paper we do not pursue them further. W → P W for the powerset functor P (which maps a set W to the set P W of subsets of W ) are exactly Kripke frames. Indeed, a map W → P W equivalently encodesa binary relation R on W . More importantly, the category theoretic notion ofa coalgebra morphism coincides with the notion of bounded (or p-) morphismin modal logic, and the coalgebraic notion of bisimulation coincides with thenotion in modal logic. This observation generalises easily to Kripke models overa set AtProp of atomic propositions and with multiple relations indexed by a setof agents Ag , which are exactly coalgebras W → ( P W ) Ag × AtProp . As shown by [3], one can construct a ‘universal model’ Z by taking the disjointunion of all coalgebras M and quotienting by bisimilarity. This coalgebra Z is final , that is, for any coalgebra M there is exactly one morphism M → Z . Theproperty of finality characterises Z up to isomorphism. Z may be a proper class. In [3], any functor F on sets is extended toclasses and it is shown that the extended functor always has a final coalgebra,constructed as the bisimilarity collapse of the disjoint union of all coalgebras. In[13], the same construction is recast in terms of an inaccessible cardinal, stayinginside the set-theoretic universe without using classes. In [5] these results aregeneralized from sets to other similar categories such as posets, and in [4], it isshown that any functor F on classes is the extension of a functor F on sets. Z classifies bisimilarity. The importance of the theorems above is notmerely the existence of the final coalgebra. Since all of these theorems involvetwo functors, one on ‘large’ sets extending another one on ‘small’ sets, andsince one is interested in the notion of bisimilarity associated with the smallfunctor, the existence of a final coalgebra for the large functor is not in itselfthe result one is interested in. But it is a fact, expressed for example as thesmall subcoalgebra lemma in [3], that in all of the constructions above, the finalcoalgebra for the large functor classifies the notion of bisimilarity associatedwith the small functor. In other words, passing from small to large does notextend—up to bisimilarity—the range of available models.
Frame conditions on Z . Often, one is interested in Kripke models satisfyingadditional frame conditions such as reflexivity, transitivity, equivalence, etc. Asufficient condition for the existence of a final coalgebra under such additionalconditions is that these conditions can be formulated by modal axioms or rules,see [34, 33] for details. 29 .2 Final coalgebra semantics of modal logic
Summing up the discussion in the previous subsection, there is a one-to-onecorrespondence between subsets of the final coalgebra and unary predicatesinvariant under bisimilarity. Therefore, whenever we know that A is a formulainvariant under bisimilarity, we may declare the subset [[ A ]] Z = { z ∈ Z | Z , z (cid:13) A } of the final coalgebra as the (final) semantics of A and recover [[ A ]] M ⊆ W as [[ A ]] M = f − ([[ A ]] Z ) , (10)where f is the unique homomorphism f : M → Z provided by the property of Z being final. Let us note that this approach isquite general: it only needs a notion of bisimilarity tied to the morphisms ofsome category (see [37] for a general definition) and a notion of modal formulawhose semantics is invariant under this notion of bisimilarity. Final coalgebra semantics of dynamic modalities.
Dynamic logics addto Kripke semantics a facility for updating the Kripke model interpreting aformula. Typically, despite seemingly increasing the expressiveness of modallogic, such dynamic logics also enjoy bisimulation invariance and can thereforebe interpreted in the final coalgebra.Whereas the Kripke semantics of an action α is a relation between pointedmodels, the final coalgebra semantics of an action α is simply a relation onthe carrier Z of the final coalgebra Z . The precise relationship between Kripkesemantics and final coalgebra semantics of actions is as follows. Let us write z α Z z ′ to express that the two points z, z ′ of the final coalgebra are related by α ,formalising that in z the action α can happen and has z ′ as a successor. Then z α Z z ′ iff there are pointed models ( M , w ) and ( M ′ , w ′ ) related by the action α such that the unique morphisms M → Z and M ′ → Z map w to z and w ′ to z ′ . Specific desiderata for epistemic actions.
The specific feature of epi-stemic actions versus arbitrary actions is that epistemic actions do not changethe factual states of affairs. Semantically, this motivates the additional require-ment that if α Z ⊆ Z × Z is the interpretation of an epistemic action α and z, z ′ ∈ Z are such that zα Z z ′ , then { p ∈ AtP rop | z (cid:13) p } = { p ∈ AtP rop | z ′ (cid:13) p } . Adjoints of dynamic modalities.
To semantically justify the full displayproperty of display calculi for dynamic logics, adjoints need to be available notonly for the standard modalities, but also for the dynamic ones. Now, it is wellknown that modalities induced by a relation come in adjoint pairs. Let us recall30 roposition 2.
Every relation R ⊆ X × Y gives rise to the modal operators h R i , [ R ] : P Y → P X and h R ◦ i , [ R ◦ ] : P X → P Y defined as follows: for every V ⊆ X and every U ⊆ Y , h R i U = { x ∈ X | ∃ y . xRy & y ∈ U } [ R ] U = { x ∈ X | ∀ y . xRy ⇒ y ∈ U }h R ◦ i V = { y ∈ Y | ∃ x . xRy & x ∈ V } [ R ◦ ] V = { y ∈ Y | ∀ x . xRy ⇒ x ∈ V } .These operators come in adjoint pairs: h R i U ⊆ V iff U ⊆ [ R ◦ ] V (11) h R ◦ i V ⊆ U iff V ⊆ [ R ] U. (12)In order to apply this proposition to dynamic modalities, we need to considerthe relation corresponding to an action α . Kripke semantics suggests to consider α as a relation on all pointed Kripke models ( M , w ) , but this would introducea two-tiered semantics: with the semantics of an ordinary modality given by arelation on the carrier of a model M and the semantics of a dynamic modalitygiven by a relation on the set of all pointed models ( M , w ) . In the final coalgebrasemantics all relations are relations on the final coalgebra Z and we can directlyapply the above proposition to both static and dynamic modalities (with the X and Y of the proposition being the carrier of the final coalgebra). Soundness of the display postulates.
Let us expand on how to interpretdisplay-type structures and sequents in the final coalgebra. Structures will betranslated into formulas, and formulas will be interpreted as subsets of the finalcoalgebra. In order to translate structures as formulas, structural connectivesneed to be translated as logical connectives; to this effect, structural connectivesare associated with pairs of logical connectives and any given occurrence of astructural connective is translated as one or the other, according to which sideof the sequent the given occurrence can be displayed on as main connective, asreported in Table 1. These logical connectives in turn are interpreted in thefinal coalgebra in the standard way. For example, [[ h α i A ]] Z = h α Z i [[ A ]] Z [[[ α ] A ]] Z = [ α Z ][[ A ]] Z [[ h α i A ]] Z = h α Z ◦ i [[ A ]] Z [[ [ α ] A ]] Z = [ α Z ◦ ][[ A ]] Z where the notation on the right-hand sides refers to the one defined in Propos-ition 2.Sequents A ⊢ B will be interpreted as inclusions [[ A ]] Z ⊆ [[ B ]] Z ; rules ( A i ⊢ B i | i ∈ I ) /C ⊢ D will be interpreted as implications of the form “if [[ A i ]] Z ⊆ [[ B i ]] Z for every i ∈ I , then [[ C ]] Z ⊆ [[ D ]] Z ”. As a direct consequence of the adjunctions(11) and (12), the following display postulates are sound under the interpretationabove. { α } X ⊢ Y { α } { α } X ⊢ { α } Y X ⊢ { α } Y { α } { α } { α } X ⊢ Y ⊤ ⊥ A ; B A ∧ B A ∨ BA > B A ∧ B A → B { a } A h a i A [ a ] A { a } A h a i A [ a ] A { α } A h α i A [ α ] A { α } A h α i A [ α ] A Table 1: Translation of structural connectives into logical connectives ✚✙✛✘ u p, r ⑦ ✚✙✛✘ u p, r ✚✙✛✘ v q q✐⑦ ❂ Figure 1: The models M α and M . Remark.
On the other hand, standard Kripke models are not in general closedunder (the interpretations of) α and α ◦ . As a direct consequence of this fact,we can show that e.g. the display postulate (cid:0) { α } { α } (cid:1) is not sound if we interpret itin a Kripke model M for any interpretation of formulas of the form [ α ] B in M .Indeed, consider the model M represented on the right-hand side of the Figure 1and let the action α be so that updating ( M , u ) gives the model M α depicted onthe left-hand side of the figure. In other words, α is the public announcement(cf. [12]) of the atomic proposition r . Further, let A := [ a ] p and B := q , where a is the agent whose equivalence relation is depicted by the arrows of the figure.Let i : M α ֒ → M be the submodel injection map. Clearly, [[[ a ] p ]] M = ∅ , whichimplies that the inclusion [[ A ]] M ⊆ [[ [ α ] B ]] M trivially holds for any interpretationof [ α ] B in M ; however, i [[[[ a ] p ]] M α ] = { u } , hence [[ h α i [ a ] p ]] M = V ( r ) ∩ { u } = { u } 6⊆ { v } = [[ q ]] M , which falsifies the inclusion [[ h α i A ]] M ⊆ [[ B ]] M . This provesour claim. Related work.
Final coalgebra semantics for dynamic logics was employedby Gerbrandy and Groeneveld [27], Gerbrandy [26], Baltag [10], and Cîrsteaand Sadrzadeh [17]. Adjoints of dynamic modalities with Kripke semanticswere considered in Baltag, Coecke, Sadrzadeh [11]. To guarantee the soundnessof the rules involving the adjoints, they have to close the Kripke models underactions, which amounts, from our point of view, to generating a subcoalgebraof the final coalgebra closed under actions. The arguments reported here infavour of the final coalgebra semantics for treating dynamic modalities withtheir adjoints are taken from [30]. 32
Proof-Theoretic Semantics for EAK
In the present section, we introduce the calculus D’.EAK for the logic EAK,which is a revised and improved version of the calculus D.EAK discussed in sec-tion 4.4. We argue that D’.EAK satisfies the requirements discussed in section2.3. On the basis of this, we propose D’.EAK as an adequate calculus from theviewpoint of proof-theoretic semantics. We also verify that D’.EAK is a quasiproper display calculus (cf. definition 2), and hence its cut elimination theoremfollows from theorem 1.
As is typical of display calculi, D’.EAK manipulates sequents of type X ⊢ Y ,where X and Y are structures, i.e. syntactic objects inductively built fromformulas using structural connectives , or proxies . Every proxy is typically asso-ciated with two logical (operational) connectives, and is interpreted contextually as one or the other of the two, depending on whether it occurs in precedent orin succedent position (cf. definition 1). The design of D’.EAK follows Došen’sprinciple (cf. section 2.3); consequently, D’.EAK is modular along many dimen-sions. For instance, the space of the versions of EAK on nonclassical bases, downto e.g. the Lambek calculus, can be captured by suitably removing structuralrules. Moreover, also w.r.t. static modal logic, the space of properly displayablenormal modal logics (cf. [32]) can be reconstructed by adding or removing struc-tural rules in a suitable way. Finally, different types of interaction between thedynamic and the epistemic modalities can be captured by changing the relativestructural rules.In order to highlight this modularity, we will present the system piecewise.First we give rules for the propositional base, divided into structural rules andoperational rules; then we do the same for the static modal operators; finally,we introduce the rules for the dynamic modalities.In the table below, we give an overview of the logical connectives of thepropositional base and their proxies. Structural symbols < > ; I Operational symbols ∧ ← ∧ → ∧ ∨ ⊤ ⊥ The table below contains the structural rules for the propositional base:33 tructural Rules
Id p ⊢ p X ⊢ A A ⊢ Y CutX ⊢ YX ⊢ Y I L I ⊢ Y < X X ⊢ Y I R X < Y ⊢ I X ⊢ Y I L I ⊢ X > Y X ⊢ Y I R Y > X ⊢ II ⊢ X I W L Y ⊢ X X ⊢ I I W R X ⊢ YX ⊢ ZW L Y ⊢ Z < X X ⊢ Z W R X < Z ⊢ YX ⊢ ZW L Y ⊢ X > Z X ⊢ Z W R Z > X ⊢ YX ; X ⊢ YC L X ⊢ Y Y ⊢ X ; X C R Y ⊢ XY ; X ⊢ ZE L X ; Y ⊢ Z Z ⊢ X ; Y E R Z ⊢ Y ; XX ; ( Y ; Z ) ⊢ WA L ( X ; Y ) ; Z ⊢ W W ⊢ ( Z ; Y ) ; X A R W ⊢ Z ; ( Y ; X ) The top-to-bottom direction of each I-rule is a special case of the corres-ponding weakening rule. However, we state them all the same for the sake ofmodularity, since they might still be part of a calculus for a substructural logicwithout weakening. The weakening rules are not given in the usual shape; thepresent version has the advantage that the new structure is introduced in isola-tion; nevertheless, the standard version is derivable from the display postulates,as shown below: X ⊢ ZY ⊢ Z < XY ; X ⊢ Z Having both versions of weakening as primitive rules is useful for reducing thesize of derivations. In the following table, we include the display postulateslinking the structural connective ; with > and < : Display Postulates X ; Y ⊢ Z (; , < ) X ⊢ Z < Y Z ⊢ X ; Y ( <, ; ) Z < Y ⊢ XX ; Y ⊢ Z (; , > ) Y ⊢ X > Z Z ⊢ X ; Y ( >, ; ) X > Z ⊢ Y In the current presentation, more connectives with their associated rules are ac-counted for than in [30]. The additional rules can be proved to be derivable fromthe remaining ones in the presence of the rules exchange E L and E R . Likewise,as is well known, by dispensing with contraction , weakening and associativity ,34n even wider array of connectives would ensue (for instance, dispensing with weakening and contraction would separate the additive and the multiplicativeversions of each connective, etc.). We are not going to expand on these wellknown ideas any further, but only point out that, in the context of the wholesystem that we are going to introduce below, this would give a modular accountof different versions of EAK with different substructural logics as propositionalbase. The calculus introduced here is amenable to this line of investigation. Anatural question in this respect would be to relate these ensuing proof formal-isms with the semantic settings of [11].In line with this modular perspective on the propositional base for EAK,the classical base is obtained by adding the so-called Grishin rules (followinge.g. [29]), encoding validities which are classical but not intuitionistic:
Grishin rules
X > ( Y ; Z ) ⊢ WGri L ( X > Y ); Z ⊢ W W ⊢ X > ( Y ; Z ) Gri R W ⊢ ( X > Y ); Z This modular treatment can be regarded as an application of Došen’s principle:calculi for versions of EAK with stronger and stronger propositional bases areobtained by progressively adding structural rules, but keeping the same oper-ational rules. As a consequence, cut elimination for the different versions willfollow immediately from the cut-elimination metatheorem without having toverify condition C again.The following table shows the operational rules for the propositional base: Operational Rules ⊥ L ⊥ ⊢ I X ⊢ I ⊥ R X ⊢ ⊥ I ⊢ X ⊤ L ⊤ ⊢ X ⊤ R I ⊢ ⊤ A ; B ⊢ Z ∧ L A ∧ B ⊢ Z X ⊢ A Y ⊢ B ∧ R X ; Y ⊢ A ∧ BA ⊢ X B ⊢ Y ∨ L A ∨ B ⊢ X ; Y Z ⊢ A ; B ∨ R Z ⊢ A ∨ BB ⊢ Y X ⊢ A ← L B ← A ⊢ Y < X Z ⊢ B < A ← R Z ⊢ B ← AB < A ⊢ Z ∧ L B ∧ A ⊢ Z Y ⊢ B A ⊢ X ∧ R Y < X ⊢ B ∧ AX ⊢ A B ⊢ Y → L A → B ⊢ X > Y Z ⊢ A > B → R Z ⊢ A → BA > B ⊢ Z ∧ L A ∧ B ⊢ Z A ⊢ X Y ⊢ B ∧ R X > Y ⊢ A ∧ B As is well known, in the presence of exchange , the connectives ← and ∧ areidentified with → and ∧ , respectively. Notice that the rules ⊥ R and ⊤ L are35erivable in the presence of weakening and the I-rules. An example of such aderivation is given below: X ⊢ II > X ⊢ ⊥ X ⊢ I ; ⊥ X < ⊥ ⊢ I X ⊢ ⊥ The rules for the normal epistemic modalities can be added to the systemabove or to any of its variants discussed early on. To this end, the language isnow expanded with two contextual proxies and four operational connectives forevery agent a , as follows: Structural symbols { a } { a } Operational symbols h a i [ a ] h a i [ a ] The proxies { a } and { a } are translated into diamond-type modalities when oc-curring in precedent position and into box-type modalities when occurring insuccedent position. The structural rules, the display postulates, and the opera-tional rules for the static modalities are respectively given in the following threetables: Structural Rules I ⊢ Xnec epL { a } I ⊢ X X ⊢ I nec epR X ⊢ { a } II ⊢ X ep nec L { a } I ⊢ X X ⊢ I ep nec R X ⊢ { a } I { a } Y > { a } Z ⊢ XF S epL { a } ( Y > Z ) ⊢ X Y ⊢ { a } X > { a } Z F S epR Y ⊢ { a } ( X > Z ) { a } X ; { a } Y ⊢ Zmon epL { a } ( X ; Y ) ⊢ Z Z ⊢ { a } Y ; { a } X mon epR Z ⊢ { a } ( Y ; X ) { a } Y > { a } X ⊢ Z ep F S L { a } ( Y > X ) ⊢ Z Y ⊢ { a } X > { a } Z ep F S R Y ⊢ { a } ( X > Z ) { a } X ; { a } Y ⊢ Z ep mon L { a } ( X ; Y ) ⊢ Z Z ⊢ { a } Y ; { a } X ep mon R Z ⊢ { a } ( Y ; X ) Notice that the mon -rules (the soundness of which is due to the monotonicityof h α i and [ α ] ) are derivable from the F S -rules in the presence of non restricted weakening and contraction .The
F S -rules above encode the following Fischer Servi-type axioms: h a i A → [ a ] B ⊢ [ a ]( A → B ) h a i A → [ a ] B ⊢ [ a ] ( A → B ) h a i ( A ∧ B ) ⊢ [ a ] A ∧ h a i B h a i ( A ∧ B ) ⊢ [ a ] A ∧ h a i B. h a i and [ a ] (and h a i and [ a ] ), namely,that they are interpreted semantically using the same relation in a Kripke frame.This link can be alternatively expressed by conjugation axioms , given below bothin the diamond- and in the box-version: h a i A ∧ B ⊢ h a i ( A ∧ h a i B ) h a i A ∧ B ⊢ h a i ( A ∧ h a i B ) , (13) [ a ]( [ a ] A ∨ B ) ⊢ ( A ∨ [ a ] B ) [ a ] ([ a ] A ∨ B ) ⊢ ( A ∨ [ a ] B ) , (14)which in turn can be encoded in the following conjugation rules: { a } ( X ; { a } Y ) ⊢ Zconj { a } X ; Y ⊢ Z X ⊢ { a } ( Y ; { a } Z ) conjX ⊢ { a } Y ; Z { a } ( X ; { a } Y ) ⊢ Zconj { a } X ; Y ⊢ Z X ⊢ { a } ( Y ; { a } Z ) conjX ⊢ { a } Y ; Z The conj -rules and the
F S -rules can be shown to be interderivable thanks tothe following display postulates.
Display Postulates { a } X ⊢ Y ( { a } , { a } ) X ⊢ { a } Y X ⊢ { a } Y ( { a } , { a } ) { a } X ⊢ Y Operational Rules { a } A ⊢ X h a i L h a i A ⊢ X X ⊢ A h a i R { a } X ⊢ h a i AA ⊢ X [ a ] L [ a ] A ⊢ { a } X X ⊢ { a } A [ a ] R X ⊢ [ a ] A { a } A ⊢ X h a i L h a i A ⊢ X X ⊢ A h a i R { a } X ⊢ h a i AA ⊢ X [ a ] L [ a ] A ⊢ { a } X X ⊢ { a } A [ a ] R X ⊢ [ a ] A The rules presented so far are essentially adaptations of display calculi of Goré’s[29]. Let us turn to the dynamic part of the calculus D’.EAK: the language isnow expanded by adding, for each action α :- two contextual proxies, together with their four corresponding operationalunary connectives;- one constant symbol and its corresponding structural proxy: Structural symbols { α } { α } Φ α Operational symbols h α i [ α ] h α i [ α ] α
37s in the previous version D.EAK, the proxies { α } and { α } are translated intodiamond-type modalities when occurring in precedent position, and into box-type modalities when occurring in succedent position. An important differencebetween D.EAK and D’.EAK is the introduction of the structural and opera-tional constants Φ α and α ; indeed, the additional expressivity they provide isused to capture the proof-theoretic behaviour of the metalinguistic abbreviation P re ( α ) at the object-level. As was the case of P re ( α ) in D.EAK, the rules belowwill be such that the proxy Φ α can occur only in precedent position. Hence, the Φ α can never be interpreted as anything else than α . However, a natural wayto extend D’.EAK would be to introduce an operational constant α , intuitivelystanding for the postconditions of α for each action α , and dualize the relevantrules so as to capture the behaviour of postconditions. In the present paper,this expansion is not pursued any further.The two tables below introduce the structural rules for the dynamic modal-ities which are analogous to those for the static modalities given early on. Structural Rules I ⊢ Xnec dynL { α } I ⊢ X X ⊢ I nec dynR X ⊢ { α } II ⊢ X dyn nec L { α } I ⊢ X X ⊢ I dyn nec R X ⊢ { α } I { α } Y > { α } Z ⊢ XF S dynL { α } ( Y > Z ) ⊢ X Y ⊢ { α } X > { α } Z F S dynR Y ⊢ { α } ( X > Z ) { α } X ; { α } Y ⊢ Zmon dynL { α } ( X ; Y ) ⊢ Z Z ⊢ { α } Y ; { α } X mon dynR Z ⊢ { α } ( Y ; X ) { α } Y > { α } X ⊢ Z dyn F S L { α } ( Y > X ) ⊢ Z Y ⊢ { α } X > { α } Z dyn F S R Y ⊢ { α } ( X > Z ) { α } X ; { α } Y ⊢ Z dyn mon L { α } ( X ; Y ) ⊢ Z Z ⊢ { α } Y ; { α } X dyn mon R Z ⊢ { α } ( Y ; X ) Analogous considerations as those made for the epistemic
F S - and mon -rulesapply to the dynamic
F S - and mon -rules above, also in relation to analogous conjugation rules.
Display Postulates { α } X ⊢ Y ( { α } , { α } ) X ⊢ { α } Y Y ⊢ { α } X ( { α } , { α } ) { α } Y ⊢ X Next, we introduce the structural rules which are to capture the specificbehaviour of epistemic actions
Atom atom Γ p ⊢ ∆ p Γ and ∆ are arbitrary finite sequences of the form ( α ) . . . ( α n ) , such thateach ( α j ) is of the form { α j } or of the form { α j } , for ≤ j ≤ n . Intuitively, the atom rules capture the requirement that epistemic actions do not change thefactual state of affairs (in the Hilbert-style presentation of EAK, this is encodedin the axiom (1) in section 4.1). Structural Rules for Epistemic Actions X ⊢ Y balance { α } X ⊢ { α } Y { α } { α } X ⊢ Ycomp αL Φ α ; X ⊢ Y X ⊢ { α } { α } Y comp αR X ⊢ Φ α > Y Φ α ; { α } X ⊢ Y reduce’ L { α } X ⊢ Y Y ⊢ Φ α > { α } X reduce’ R Y ⊢ { α } X { α }{ a } X ⊢ Y swap-in’ L Φ α ; { a }{ β } α a β X ⊢ Y Y ⊢ { α }{ a } X swap-in’ R Y ⊢ Φ α > { a }{ β } α a β X (cid:16) { a }{ β } X ⊢ Y | α a β (cid:17) swap-out’ L { α }{ a } X ⊢ ; (cid:16) Y | α a β (cid:17) (cid:16) Y ⊢ { a }{ β } X | α a β (cid:17) swap-out’ R ; (cid:16) Y | α a β (cid:17) ⊢ { α }{ a } X The swap-in’ rules are unary and should be read as follows: if the premise holds,then the conclusion holds relative to any action β such that α a β . The swap-out’ rules do not have a fixed arity; they have as many premises as there areactions β such that α a β . In the conclusion, the symbol ; (cid:16) Y | α a β (cid:17) refers toa string ( · · · ( Y ; Y ) ; · · · ; Y ) with n occurrences of Y , where n = |{ β | α a β }| .The swap-in and swap-out rules encode the interaction between dynamic andepistemic modalities as it is captured by the interaction axioms in the Hilbertstyle presentation of EAK (cf. (4) in section 4.1 and similarly in section 4.2). The reduce rules encode well-known EAK validities such as h α i A → ( P re ( α ) ∧h α i A ) .Finally, the operational rules for h α i , [ α ] , and α are given in the table below: Operational Rules { α } A ⊢ X h α i L h α i A ⊢ X X ⊢ A h α i R { α } X ⊢ h α i AA ⊢ X [ α ] L [ α ] A ⊢ { α } X X ⊢ { α } A [ α ] R X ⊢ [ α ] A Φ α ⊢ X αL α ⊢ X αR Φ α ⊢ α Soundness.
The calculus D’.EAK can be readily shown to be sound with re-spect to the final coalgebra semantics. The general procedure has been outlined The swap-out rule could indeed be infinitary if action structures were allowed to be infinite,which in the present setting, as in [12], is not the case.
39n section 5. The soundness of most of the rules of D’.EAK can be shown en-tirely analogously to the soundness of the corresponding rules in D.EAK, whichis outlined in [30].As for rules not involving { α } , we will rely on the following observation, whichis based on the invariance of EAK-formulas under bisimulation (cf. Section 4.1): Lemma 3.
The following are equivalent for all EAK-formulas A and B :(1) [[ A ]] Z ⊆ [[ B ]] Z ;(2) [[ A ]] M ⊆ [[ B ]] M for every model M .Proof. The direction from (2) to (1) is clear; conversely, fix a model M , and let f : M → Z be the unique arrow; then (1) immediately implies that [[ A ]] M = f − ([[ A ]] Z ) ⊆ f − ([[ B ]] Z ) = [[ B ]] M .In the light of the lemma above, and using the translations provided in Table1, the soundness of unary rules A ⊢ B/C ⊢ D not involving { α } , such as balance , h α i R and [ α ] L , can be straightforwardly checked as implications of the form “if [[ A ]] M ⊆ [[ B ]] M on every model M , then [[ C ]] M ⊆ [[ D ]] M on every model M ”. Asan example, let us check the soundness of balance : Let A, B be EAK-formulassuch that [[ A ]] M ⊆ [[ B ]] M on every model M . Let us fix a model M , and showthat [[ h α i A ]] M ⊆ [[[ α ] B ]] M . As discussed in [36, Subsection 4.2], the followingidentities hold in any standard model: [[ h α i A ]] M = [[ P re ( α )]] M ∩ ι − k [ i [[[ A ]] M α ]] , (15) [[[ α ] A ]] M = [[ P re ( α )]] M ⇒ ι − k [ i [[[ A ]] M α ]] , (16)where the map i : M α → ` α M is the submodel embedding, and ι k : M → ` α M is the embedding of M into its k -colored copy. Letting g ( − ) := ι − k [ i [ − ]] ,we need to show that [[ P re ( α )]] M ∩ g ([[ A ]] M α ) ⊆ [[ P re ( α )]] M ⇒ g ([[ B ]] M α ) . This is a direct consequence of the Heyting-valid implication “if b ≤ c then a ∧ b ≤ a → c ”, the monotonicity of g , and the assumption that [[ A ]] M ⊆ [[ B ]] M holds on every model, hence on M α .Actually, for all rules ( A i ⊢ B i | i ∈ I ) /C ⊢ D not involving { α } except balance , h α i R and [ α ] L , stronger soundness statements can be proven of the form“for every model M , if [[ A i ]] M ⊆ [[ B i ]] M for every i ∈ I , then [[ C ]] M ⊆ [[ D ]] M ”(this amounts to the soundness w.r.t. the standard semantics). This is the casefor all display postulates not involving { α } , the soundness of which boils downto the well known adjunction conditions holding in every model M . As to theremaining rules not involving { α } , thanks to the following general principle of indirect (in)equality , the stronger soundness condition above boils down to theverification of inclusions which interpret validities of IEAK [36], and hence, afortiori, of EAK. Same arguments hold for the Grishin rules, except that theirsoundness boils down to classical but not intuitionistic validities.40 emma 4. (Principle of indirect inequality) Tfae for any preorder P andall a, b ∈ P :(1) a ≤ b ;(2) x ≤ a implies x ≤ b for every x ∈ P ;(3) b ≤ y implies a ≤ y for every y ∈ P . As an example, let us verify s-out L : fix a model M , fix EAK-formulas A and B , and assume that for every action β , if α a β then [[ h a ih β i A ]] M ⊆ [[ B ]] M , i.e.,that S { [[ h a ih β i A ]] M | α a β } ⊆ [[ B ]] M ; we need to show that [[ h α ih a i A ]] M ⊆ [[ B ]] M .By the principle of indirect inequality, it is enough to show that [[ h α ih a i A ]] M ⊆ S { [[ h a ih β i A ]] M | α a β } . Indeed, since axiom (4) is valid on any model, we have: [[ h α ih a i A ]] M ⊆ [[ P re ( α )]] M ∩ [ { [[ h a ih β i A ]] M | α a β } ⊆ [ { [[ h a ih β i A ]] M | α a β } . The soundness of the operational rules of α is immediate; the soundness of atom can be proven directly on the final coalgebra by induction on the lengthof Γ and ∆ using the fact, mentioned on page 30, that epistemic actions do notchange the valuations of atomic formulae. For instance, as to the base case ofthis induction, let us argue for the soundness of { α } p ⊢ p and p ⊢ { α } p : indeed,let α Z ⊆ Z × Z be the interpretation of the epistemic action α on the finalcoalgebra, then the left-hand side of the atom-sequent above is interpreted as theset α Z [[[ p ]] Z ] . Because of the assumption on α Z mentioned above it immediatelyfollows that α Z [[[ p ]] Z ] ⊆ [[ p ]] Z , and α Z [[[ p ]] cZ ] ⊆ [[ p ]] cZ . The former inclusion givesthe soundness of { α } p ⊢ p , while the latter is equivalent to [[ p ]] Z ⊆ ( α Z [[[ p ]] cZ ]) c ,which gives the soundness of p ⊢ { α } p .The soundness of the comp rules is given in the appendix (cf. subsectionA.2).Finally, the soundness of the rules which do involve { α } remains to be shown.The soundness of the display postulates immediately follows from Proposition 2.As an example, let us verify the soundness of dyn F S L : translating the structuresinto formulas, it boils down to verifying that, for all EAK-formulas A, B and C ,if [[ [ α ] A ]] Z ∧ [[ h α i B ]] Z ⊆ [[ C ]] Z , then [[ h α i ( A ∧ B )]] Z ⊆ [[ C ]] Z . By applying theappropriate adjunction rules, the implication above is equivalent to the followingimplication: if [[ B ]] Z ⊆ [[[ α ]( [ α ] A ∨ C )]] Z then [[ B ]] Z ⊆ [[ A ∨ [ α ] C ]] Z . By applyingthe principle of indirect inequality, we are reduced to showing the inclusion [[[ α ]( [ α ] A ∨ C )]] Z ⊆ [[ A ∨ [ α ] C ]] Z , which is the soundness of the box-version of a conjugation condition (see theshape of (14) for epistemic modalities), and is true in Z since [ α ] is interpretedas [ α ◦ ] . Completeness and conservativity.
The completeness of D’.EAK w.r.t. theHilbert presentation of EAK (cf. subsections 4.1 and 4.2) is achieved by showingthat the axioms of (the intuitionistic version of) EAK are derivable in D’.EAK.These derivations are collected in subsection C.41gain, as was the case for D.EAK, the fact that D’.EAK is a conservativeextension of EAK can be argued as follows: let
A, B be EAK-formulas suchthat A ⊢ D ′ .EAK B , and let Z be the final coalgebra. By the soundness ofD’.EAK w.r.t. the Z , this implies that [[ A ]] Z ⊆ [[ B ]] Z , which, by the bisimulationinvariance of EAK (cf. [30, Lemma 1]), implies that [[ A ]] M ⊆ [[ B ]] M for everyKripke model M , which, by the completeness of EAK w.r.t. the standard Kripkesemantics, implies that A ⊢ EAK B . Adequacy of D’.EAK w.r.t. Wansing’s criteria.
It is easy to see that thecalculus D’.EAK enjoys the display property (cf. Definition 1). Like its previousversion, D’.EAK is defined independently of the relational semantics of EAK,and therefore is suitable for a fine-grained proof-theoretic semantic analysis. Itcan be readily verified by inspection that all operational rules satisfy Wansing’scriteria of separation , symmetry and explicitness (cf. subsection 2.3).Moreover, a clear-cut division of labour has been achieved between the op-erational rules, which are to encode the proof-theoretic meaning of the newconnectives, and the structural rules, which are to express the relations enter-tained between the different connectives by way of their proxies.Another important proof-theoretic feature of D’.EAK is modularity. Asdiscussed in subsection 6.1, by suitably removing structural rules for the pro-positional base of D’.EAK, the substructural versions of EAK can be modularlydefined. Moreover, by adding structural rules corresponding to properly dis-playable modal logics (cf. [32]), different assumptions can be captured on thebehaviour of the epistemic modalities. Notwithstanding the fact that the old reverse rules, offending segregation , arederived rules in D’.EAK, still the system D’.EAK does not satisfy segregation .However, the only rule in D’.EAK offending segregation is atom because oneof the two principal formulas in each atom axioms might not occur in display.Even if the most rigid proof-theoretic semantic principle is not met, D’.EAK is aquasi-proper display calculus, and hence it enjoys Belnap-style cut elimination,as will be shown in the next subsection. In the present subsection, we prove that D’.EAK is a quasi proper displaycalculus (cf. Subsection 3.1), that is, the rules of D’.EAK satisfy conditionsC , C , C , C , C ′ , C ′′ , C , C , C . By Theorem 1, this is enough to establishthat the calculus enjoys the cut elimination and the subformula property.The rules reverse are now derivable, and all the rules with the side condition P re ( α ) have been reformulated so as to either remove P re ( α ) altogether, or toreplace it with its structural counterpart. This has been achieved by expanding Note that
Balance , comp , reduce , swap-in and swap-out are the only specific structuralrules for epistemic actions; the monotonicity and Fischer-Servi rules respectively encode theconditions that box and diamond are monotone and interpreted by means of the same relation;the necessitation can be considered as a special case of atom and IW can be eliminated if,e.g., ⊥ ⊢ h α i⊥ and [ α ] ⊤ ⊢ ⊤ are introduced as zeroary rules. P re ( α ) can be replaced byan operational constant and its corresponding structural connective. Hence, itcan be readily verified that all rules are closed under simultaneous substitution ofarbitrary structures for congruent parameters, which satisfies conditions C andC . It is easy to see that the operational rules for α and the comp rules satisfythe criteria C –C . The atom axioms can be readily seen to verify condition C ′′ as given in subsection 3.1.Finally, as to condition C , let us show the cases involving the new connective α . All the other cases are reported in appendix B. Φ α ⊢ α ... π Φ α ⊢ X α ⊢ X Φ α ⊢ X ... π Φ α ⊢ X In the present paper, we provide an analysis, conducted adopting the view-point of proof-theoretic semantics, of the state-of-the-art deductive systems fordynamic epistemic logic, focusing mainly on Baltag-Moss-Solecki’s logic of epi-stemic actions and knowledge (EAK). We start with an overview of the generalresearch agenda in proof-theoretic semantics, and then we focus on display cal-culi, as a proof-theoretic paradigm which has been successful in accounting fordifficult logics, such as modal logics and substructural logics. We discuss therequirements which a proof system should satisfy to provide adequate proof-theoretic semantics to logical constants, and, as an original contribution, we in-troduce the notion of quasi proper display calculus, and prove its correspondingBelnap-style cut elimination metatheorem. We then evaluate the main existingproof systems for PAL/EAK according to the previously discussed requirements.As the second original contribution, we propose a revised version of one suchsystem, namely of the system D.EAK (cf. section 4.4), and we argue that ourrevised system D’.EAK adequately meets the proof-theoretic semantic require-ments for all the logical constants involved. We also show that D’.EAK is soundw.r.t. the final coalgebra semantics, complete w.r.t. EAK, of which it is a con-servative extension. These three facts together guarantee that D’.EAK exactlycaptures EAK. Finally, we verify that D’.EAK is a quasi proper display calcu-lus. Hence, the generalized metatheorem applies, and D’.EAK is thus shown toenjoy Belnap-style cut elimination (which was not argued for in the case of theoriginal system D.EAK) and the subformula property. The main ingredient ofthis revision is an expansion of the language of the original system, aimed atachieving an independent proof-theoretic account of the preconditions
P re ( α ) .This account is independent both in the sense that it is given purely in terms ofthe resources of D’.EAK, and in the sense that the metalinguistic abbreviation P re ( α ) is treated as a first-class citizen of the revised system. Indeed, P re ( α )
43s endowed with both an operational and a structural representation, both ofwhich well-behaving.
Uniform proof-theoretic account for dynamic logics.
The presentpaper is part of a larger research program aimed at giving a uniform proof-theoretic account to a wide class of logics which includes dynamic logics. In[20] and [22], this treatment has been extended to monotone modal logics and,respectively, to the full language of Propositional Dynamic Logic. Anotherinteresting case study is Parikh’s Game Logic [43], where the dynamic modalitiesare non normal and the set of agents is endowed with algebraic structure, whichis treated in a paper [21] in preparation.
Multi-type display-style calculi.
The metatheorem proven in the presentpaper applies to a class of display calculi (the quasi-proper display calculi) whichgeneralize Wansing’s notion of proper display calculi by relaxing the propertyof isolation. However, in both quasi proper and proper display calculi, rules arerequired to be closed under simultaneous substitution of arbitrary structuresfor congruent formulas. This requirement occurs in a weaker form in both theoriginal [15, Theorem 4.4] and in some of its subsequent versions [16, 45, 49].Indeed, these metatheorems apply to display calculi admitting rules for whichthe closure under substitution may be not arbitrary , but restricted to structuressatisfying certain conditions. This weaker requirement primarily concerns rules;however, it is encoded in the notion of regular formula and asks every formula tobe regular. The condition given in terms of regular formulas is key to accountingfor important logics such as linear logic. On the other hand, it ingeniously relieson very special features of the signature of linear logic, and hence it is of difficultapplication outside that setting. We conjecture that logics such as linear logiccan be alternatively accounted for by display-type calculi all the rules of whichare closed under simultaneous substitution of arbitrary structures for parametricoperational terms (formulas). We conjecture that this is possible thanks tothe introduction of a suitable multi-type environment, in which every derivablesequent/consecution is required to be type-uniform (i.e., both the antecedent andthe consequent of any sequent/consecution must belong to the same type). Therequirement formulated in terms of regular formulas would then be encodedin the multi-type setting in terms of the condition that, in each given rule,parametric constituents (of a given and unambiguously determined type) canbe uniformly replaced by structures which are arbitrary within that same type ,so as to obtain instances of the same rule. An example of such a multi-typeenvironment is introduced in [23]. The adaptation of the multi-type setting tothe case of linear logic is work in progress.44
Special rules
A.1 Derived rules in D’.EAK
In the presence of the display postulates, the conj -rules are interderivable withthe Fischer Servi rules. Indeed, let us show that the following rules { α } ( X ; { α } Y ) ⊢ Zconj { α } X ; Y ⊢ Z Y ⊢ { α } X > { α } Z F SY ⊢ { α } ( X > Z ) are interderivable: { α } ( X ; { α } Y ) ⊢ ZX ; { α } Y ⊢ { α } Z { α } Y ; X ⊢ { α } ZX ⊢ { α } Y > { α } Z F SX ⊢ { α } ( Y > Z ) { α } X ⊢ Y > ZY ; { α } X ⊢ Z { α } X ; Y ⊢ Z Y ⊢ { α } X > { α } Z { α } X ; Y ⊢ { α } ZY ; { α } X ⊢ { α } Z { α } ( Y ; { α } X ) ⊢ Zconj { α } Y ; X ⊢ ZX ; { α } Y ⊢ Z { α } Y ⊢ X > ZY ⊢ { α } ( X > Z ) Analogous derivations show that the pairs of rules in each row of the table beloware interderivable: { α } ( X ; { α } Y ) ⊢ Zconj { α } X ; Y ⊢ Z Y ⊢ { α } X > { α } Z F SY ⊢ { α } ( X > Z ) X ⊢ { α } ( Y ; { α } Z ) conjX ⊢ { α } Y ; Z { α } Y > { α } X ⊢ ZF S { α } ( Y > X ) ⊢ ZX ⊢ { α } ( Y ; { α } Z ) conjX ⊢ { α } Y ; Z { α } Y > { α } Z ⊢ XF S { α } ( Y > Z ) ⊢ X Let us show that the rules “with side conditions” in D.EAK (cf. subsection 4.4)can be derived from their corresponding rules in D’.EAK and the remainingpart of the calculus.An important benefit of the revised system is that the operational rules reverse (or more precisely their rewritings in the new notation), which were primitivein the old system, are now derivable using the new rules for Φ α and α and thenew reduce . This supports our intuition that the rules reverse do not participatein the proof-theoretic meaning of the connectives h α i and [ α ] . Note that we are using exchange, but this rule is not required if we add the corresponding
Fisher-Servi rule for the right-residuum of ‘ ; ’ and the obvious conjugation rule with ‘ X ; { α } Y ’in a reversed order. α ⊢ α α ; { α } A ⊢ X α ⊢ X < { α } A Φ α ⊢ X < { α } A Φ α ; { α } A ⊢ X { α } A ⊢ XA ⊢ { α } X [ α ] A ⊢ { α } { α } X comp [ α ] A ⊢ Φ α > X Φ α ; [ α ] A ⊢ X [ α ] A ; Φ α ⊢ X Φ α ⊢ [ α ] A > X α ⊢ [ α ] A > X [ α ] A ; 1 α ⊢ X α ; [ α ] A ⊢ X Φ α ⊢ α X ⊢ α > { α } A α ; X ⊢ { α } A α ⊢ { α } A < X Φ α ⊢ { α } A < X Φ α ; X ⊢ { α } AX ⊢ Φ α > { α } A reduce ′ X ⊢ { α } A { α } X ⊢ A { α } { α } X ⊢ h α i Acomp Φ α ; X ⊢ h α i AX ; Φ α ⊢ h α i A Φ α ⊢ X > h α i A α ⊢ X > h α i AX ; 1 α ⊢ h α i A α ; X ⊢ h α i AX ⊢ α > h α i A The old rules reduce are derivable as follows. Φ α ⊢ α α ; { α } A ⊢ X α ⊢ X < { α } A Φ α ⊢ X < { α } A Φ α ; { α } A ⊢ Xreduce ′ { α } A ⊢ X Φ α ⊢ α X ⊢ α > { α } A α ; X ⊢ { α } AX ; 1 α ⊢ { α } A α ⊢ X > { α } A Φ α ⊢ X > { α } AX ; Φ α ⊢ { α } A Φ α ; X ⊢ { α } AX ⊢ Φ α > { α } A reduce ′ X ⊢ { α } A The old swap-in rules are derivable in the revised calculus from the new swap-in rules as follows. Φ α ⊢ α α ; { α }{ a } X ⊢ Y α ⊢ Y < { α }{ a } X Φ α ⊢ Y < { α }{ a } X Φ α ; { α }{ a } X ⊢ Y reduce ′ { α }{ a } X ⊢ Y swap - in ′ Φ α ; { a }{ β } α a β X ⊢ Y Φ α ⊢ Y < { a }{ β } α a β X α ⊢ Y < { a }{ β } α a β X α ; { a }{ β } α a β X ⊢ Y Φ α ⊢ α Y ⊢ α > { α }{ a } X α ; Y ⊢ { α }{ a } X α ⊢ { α }{ a } X < Y Φ α ⊢ { α }{ a } X < Y Φ α ; Y ⊢ { α }{ a } XY ⊢ Φ α > { α }{ a } X reduce ′ Y ⊢ { α }{ a } X swap - in ′ Y ⊢ { α }{ a } XY ⊢ Φ α > { a }{ β } α a β X The old swap-out rules (translated into D’.EAK) are derivable using the new swap-out rules: 46 α ⊢ α α ; { a }{ β } X ⊢ Y | α a β α ⊢ Y < { a }{ β } X | α a β Φ α ⊢ Y < { a }{ β } X | α a β Φ α ; { a }{ β } X ⊢ Y | α a β reduce ′ { a }{ β } X ⊢ Y | α a β · · ·· · ·· · ·· · ·· · · Φ α ⊢ α α ; { a }{ β n } X ⊢ Y | α a β n α ⊢ Y < { a }{ β n } X | α a β n Φ α ⊢ Y < { a }{ β n } X | α a β n Φ α ; { a }{ β n } X ⊢ Y | α a β n reduce ′ { a }{ β n } X ⊢ Y | α a β n swap - out ′ { α }{ a } X ⊢ ; (cid:16) Y | α a β (cid:17) α ⊢ { α }{ a } X > ; (cid:16) Y | α a β (cid:17) { α }{ a } X ; 1 α ⊢ ; (cid:16) Y | α a β (cid:17) α ; { α }{ a } X ⊢ ; (cid:16) Y | α a β (cid:17) Φ α ⊢ α Y ⊢ α > { a }{ β } X | α a β α ; Y ⊢ { a }{ β } X | α a β α ⊢ { a }{ β } X | α a β < Y Φ α ⊢ { a }{ β } X | α a β < Y Φ α ; Y ⊢ { a }{ β } X | α a β Y ⊢ Φ α > { a }{ β } X | α a β reduce ′ Y ⊢ { a }{ β } X | α a β · · ·· · ·· · ·· · ·· · ·· · ·· · · Φ α ⊢ α Y ⊢ α > { a }{ β n } X | α a β n α ; Y ⊢ { a }{ β n } X | α a β n α ⊢ { a }{ β n } X | α a β n < Y Φ α ⊢ { a }{ β n } X | α a β n < Y Φ α ; Y ⊢ { a }{ β n } X | α a β n Y ⊢ Φ α > { a }{ β n } X | α a β n reduce ′ Y ⊢ { a }{ β n } X | α a β n swap - out ′ ; (cid:16) Y | α a β (cid:17) ⊢ { α }{ a } X α ⊢ ; (cid:16) Y | α a β (cid:17) > { α }{ a } X ; (cid:16) Y | α a β (cid:17) ; 1 α ⊢ { α }{ a } X ; (cid:16) Y | α a β (cid:17) ; 1 α ⊢ { α }{ a } X ; (cid:16) Y | α a β (cid:17) ⊢ α > { α }{ a } X A.2 Soundness of comp rules in the final coalgebra
We address the reader to [30] for details on the final coalgebra semantics fordynamic epistemic logic.To prove the soundness of the rules above in the final coalgebra it suffices tocheck that for every formula A , [ α ][ α − ][[ A ]] Z ⊆ [[ P re ( α ) → A ]] Z and [[ P re ( α ) ; A ]] Z ⊆ h α ih α − i [[ A ]] Z . We will make use of the following general fact:
Fact 5.
Let R be a binary relation on a set X and let R − be its converse.Then, [ Dom ( R ) × Dom ( R )] ∩ ∆ X ⊆ R ; R − , where Dom ( R ) = { x ∈ X | xRy for some y ∈ X } , and ∆ X = { ( x, x ) | x ∈ X } .Proof. Straightforward. 47 act 6.
The following comp rules: Y ⊢ { α } { α } XY ⊢ P re ( α ) > X { α } { α } X ⊢ YP re ( α ) ; X ⊢ Y are sound in the final coalgebra.Proof. h α ih α − i [[ A ]] Z = α − [ α [[[ A ]] Z ]]= ( α ; α − )[[[ A ]] Z ] ⊇ S [[[ A ]] Z ] Fact 5 = Dom ( α ) ∩ [[ A ]] Z = [[ P re ( α ) ; A ]] Z , [ α ][ α − ][[ A ]] Z = ( α − [([ α − ][[ A ]] Z ) c ]) c = ( α − [ α [[[ A ]] c Z ]]) c = (( α ; α − )[[[ A ]] c Z ]) c ⊆ ( S [[[ A ]] c Z ]) c Fact 5 = (
Dom ( α ) ∩ [[ A ]] c Z ]) c = Dom ( α ) c ∪ [[ A ]] Z = [[ P re ( α ) → A ]] Z , where S = [ Dom ( R ) × Dom ( R )] ∩ ∆ X . B Cut elimination for D’.EAK
In the present section, we report on the remaining cases for the verification ofcondition C for D’.EAK; these cases are needed already for the cut eliminationá la Gentzen for D.EAK, but do not appear in [30].First we consider the atom rule (see page 38). Γ p ⊢ p p ⊢ ∆ p Γ p ⊢ ∆ p Γ p ⊢ ∆ p Now we treat the introductions of the connectives of the propositional base (wealso treat here the cases relative to the two additional arrows ← and ∧ addedto our presentation of D.EAK): I ⊢ ⊤ ... π I ⊢ X ⊤ ⊢ X I ⊢ X ... π I ⊢ X... πX ⊢ I X ⊢ ⊥ ⊥ ⊢ I X ⊢ I ... πX ⊢ I .. π X ⊢ A ... π Y ⊢ BX ; Y ⊢ A ∧ B ... π A ; B ⊢ ZA ∧ B ⊢ ZX ; Y ⊢ Z ... π Y ⊢ B ... π X ⊢ A ... π A ; B ⊢ ZB ; A ⊢ ZA ⊢ B > ZX ⊢ B > ZB ; X ⊢ ZX ; B ⊢ ZB ⊢ X > ZY ⊢ X > ZX ; Y ⊢ Z... π Z ⊢ B ; AZ ⊢ B ∨ A ... π B ⊢ Y ... π A ⊢ XB ∨ A ⊢ Y ; XZ ⊢ Y ; X ... π Z ⊢ B ; AZ ⊢ A ; BA > Z ⊢ B ... π B ⊢ YA > Z ⊢ YZ ⊢ A ; YZ ⊢ Y ; AY > Z ⊢ A ... π A ⊢ XY > Z ⊢ XZ ⊢ Y ; X... π Y ⊢ A > BY ⊢ A → B ... π X ⊢ A ... π B ⊢ ZA → B ⊢ X > ZY ⊢ X > Z ... π X ⊢ A ... π Y ⊢ A > BA ; Y ⊢ B ... π B ⊢ ZA ; Y ⊢ ZY ; A ⊢ ZA ⊢ Y > ZX ⊢ Y > ZY ; X ⊢ ZX ; Y ⊢ ZY ⊢ X > Z... π Y ⊢ B < AY ⊢ B ← A ... π B ⊢ Z ... π X ⊢ AB ← A ⊢ Z < XY ⊢ Z < X ... π X ⊢ A ... π Y ⊢ B < AY ; A ⊢ B ... π B ⊢ ZY ; A ⊢ ZA ; Y ⊢ ZA ⊢ Z < YX ⊢ Z < YX ; Y ⊢ ZY ; X ⊢ ZY ⊢ Z < X .. π A ⊢ Y ... π Z ⊢ BY > Z ⊢ A ∧ B ... π A > B ⊢ XA ∧ B ⊢ XY > Z ⊢ X ... π Z ⊢ B ... π A > B ⊢ XB ⊢ A ; XZ ⊢ A ; XZ ⊢ X ; AX > Z ⊢ A ... π A ⊢ YX > Z ⊢ YZ ⊢ X ; YZ ⊢ Y ; XY > Z ⊢ X... π Y ⊢ B ... π A ⊢ ZY < Z ⊢ B ∧ A ... π B < A ⊢ XB ∧ A ⊢ XY < Z ⊢ X ... π Y ⊢ B ... π B < A ⊢ XB ⊢ X ; AY ⊢ X ; AY ⊢ A ; XY < X ⊢ A ... π A ⊢ ZY < X ⊢ ZY ⊢ Z ; XY ⊢ X ; ZY < Z ⊢ Y Now we turn to the part of D’.EAK with static modalities. We omit the proofsfor h a i and [ a ] , because they analogous to the transformations of h a i and [ a ] . ... π X ⊢ A { a } X ⊢ h a i A ... π { a } A ⊢ Y h a i A ⊢ Y { a } X ⊢ Y ... π X ⊢ A ... π { a } A ⊢ YA ⊢ { a } YX ⊢ { a } Y { a } X ⊢ Y... π X ⊢ { a } AX ⊢ [ a ] A ... π A ⊢ Y [ a ] A ⊢ { a } YX ⊢ { a } Y ... π X ⊢ { a } A { a } X ⊢ A ... π A ⊢ Y { a } X ⊢ YX ⊢ { a } Y The transformations of the dynamic modalities are analogous to the ones ofstatic modalities and, again, we only show them for h α i and [ α ] . ... π X ⊢ A { α } X ⊢ h α i A ... π { α } A ⊢ Y h α i A ⊢ Y { α } X ⊢ Y ... π X ⊢ A ... π { α } A ⊢ YA ⊢ { α } YX ⊢ { α } Y { α } X ⊢ Y .. π X ⊢ { α } AX ⊢ [ α ] A ... π A ⊢ Y [ α ] A ⊢ { α } YX ⊢ { α } Y ... π X ⊢ { α } A { α } X ⊢ A ... π A ⊢ Y { α } X ⊢ YX ⊢ { α } Y C Completeness of D’.EAK
To prove, indirectly, the completeness of D’.EAK it is enough to show that all theaxioms and rules of IEAK are theorems and, respectively, derived or admissiblerules of D’.EAK. Below we show the derivations of the dynamic axioms and weleave the remaining axioms and rules to the reader. h α i p ⊣⊢ α ∧ p Φ α ⊢ α { α } p ⊢ p Φ α ; { α } p ⊢ α ∧ preduce ′ { α } p ⊢ α ∧ p h α i p ⊢ α ∧ p { α } p ⊢ p { α } { α } p ⊢ h α i pcomp Φ α ; p ⊢ h α i p Φ α ⊢ h α i p < p α ⊢ h α i p < p α ; p ⊢ h α i p α ∧ p ⊢ h α i p [ α ] p ⊣⊢ α → p p ⊢ { α } p [ α ] p ⊢ { α } { α } p comp [ α ] p ⊢ Φ α > p Φ α ; [ α ] p ⊢ p Φ α ⊢ p < [ α ] p α ⊢ p < [ α ] p α ; [ α ] p ⊢ p [ α ] p ⊢ α > p [ α ] p ⊢ α → p Φ α ⊢ α p ⊢ { α } p α → p ⊢ Φ α > { α } p reduce ′ α → p ⊢ { α } p α → p ⊢ [ α ] p h α i⊤ ⊣⊢ α Φ α ⊢ α I ⊢ α < Φ α nec { α } I ⊢ α < Φ α I ⊢ { α } (1 α < Φ α ) ⊤ ⊢ { α } (1 α < Φ α ) { α }⊤ ⊢ α < Φ α Φ α ; { α }⊤ ⊢ α reduce ′ { α }⊤ ⊢ α h α i⊤ ⊢ α I ⊢ ⊤ nec { α } I ⊢ ⊤{ α } { α } I ⊢ h α i⊤ comp Φ α ; I ⊢ h α i⊤ Φ α ⊢ h α i⊤ α ⊢ h α i⊤ α ] ⊥ ⊣⊢ ¬ α ⊥ ⊢ I nec ⊥ ⊢ { α } I [ α ] ⊥ ⊢ { α } { α } I comp [ α ] ⊥ ⊢ Φ α > I Φ α ; [ α ] ⊥ ⊢ I Φ α ; [ α ] ⊥ ⊢ ⊥ Φ α ⊢ ⊥ < [ α ] ⊥ α ⊢ ⊥ < [ α ] ⊥ α ; [ α ] ⊥ ⊢ ⊥ [ α ] ⊥ ⊢ α > ⊥ [ α ] ⊥ ⊢ α → ⊥ [ α ] ⊥ ⊢ ¬ α Φ α ⊢ α ⊥ ⊢ I nec ⊥ ⊢ { α } I { α } ⊥ ⊢ I { α } ⊥ ⊢ ⊥⊥ ⊢ { α }⊥ α → ⊥ ⊢ Φ α > { α }⊥¬ α ⊢ Φ α > { α }⊥ reduce ′ ¬ α ⊢ { α }⊥¬ α ⊢ [ α ] ⊥ h α i⊥ ⊣⊢ ⊥ ⊥ ⊢ I nec ⊥ ⊢ { α } I { α }⊥ ⊢ I { α }⊥ ⊢ ⊥h α i⊥ ⊢ ⊥ ⊥ ⊢ I ⊥ ⊢ h α i⊥ [ α ] ⊤ ⊣⊢ ⊤ I ⊢ ⊤ [ α ] ⊤ ⊢ ⊤ I ⊢ ⊤ nec { α } I ⊢ ⊤ I ⊢ { α }⊤⊤ ⊢ { α }⊤⊤ ⊢ [ α ] ⊤ [ α ]( A ∧ B ) ⊣⊢ [ α ] A ∧ [ α ] B A ⊢ AA ; B ⊢ AA ∧ B ⊢ A [ α ]( A ∧ B ) ⊢ { α } A [ α ]( A ∧ B ) ⊢ [ α ] A B ⊢ BA ; B ⊢ BA ∧ B ⊢ B [ α ]( A ∧ B ) ⊢ { α } B [ α ]( A ∧ B ) ⊢ [ α ] B [ α ]( A ∧ B ) ; [ α ]( A ∧ B ) ⊢ [ α ] A ∧ [ α ] B [ α ]( A ∧ B ) ⊢ [ α ] A ∧ [ α ] B A ⊢ A [ α ] A ⊢ { α } A { α } [ α ] A ⊢ A B ⊢ B [ α ] B ⊢ { α } B { α } [ α ] B ⊢ B { α } [ α ] A ; { α } [ α ] B ⊢ A ∧ Bmon { α } ([ α ] A ; [ α ] B ) ⊢ A ∧ B [ α ] A ; [ α ] B ⊢ { α } ( A ∧ B )[ α ] A ; [ α ] B ⊢ [ α ]( A ∧ B )[ α ] A ∧ [ α ] B ⊢ [ α ]( A ∧ B ) α i ( A ∧ B ) ⊣⊢ h α i A ∧ h α i B A ⊢ AA ; B ⊢ AA ∧ B ⊢ A { α } A ∧ B ⊢ h α i A h α i ( A ∧ B ) ⊢ h α i A B ⊢ BA ; B ⊢ BA ∧ B ⊢ B { α } A ∧ B ⊢ h α i B h α i ( A ∧ B ) ⊢ h α i B h α i ( A ∧ B ) ; h α i ( A ∧ B ) ⊢ h α i A ∧ h α i B h α i ( A ∧ B ) ⊢ h α i A ∧ h α i B A ⊢ A balance { α } A ⊢ { α } A { α } { α } A ⊢ A B ⊢ B balance { α } B ⊢ { α } B { α } { α } B ⊢ B { α } { α } A ; { α } { α } B ⊢ A ∧ B mon { α } ( { α } A ; { α } B ) ⊢ A ∧ B { α } { α } ( { α } A ; { α } B ) ⊢ h α i ( A ∧ B ) comp Φ α ; ( { α } A ; { α } B ) ⊢ h α i ( A ∧ B )(Φ α ; { α } A ) ; { α } B ⊢ h α i ( A ∧ B )Φ α ; { α } A ⊢ h α i ( A ∧ B ) < { α } Breduce ′ { α } A ⊢ h α i ( A ∧ B ) < { α } B h α i A ⊢ h α i ( A ∧ B ) < { α } B h α i A ; { α } B ⊢ h α i ( A ∧ B ) { α } B ⊢ h α i A > h α i ( A ∧ B ) h α i B ⊢ h α i A > h α i ( A ∧ B ) h α i A ; h α i B ⊢ h α i ( A ∧ B ) h α i A ∧ h α i B ⊢ h α i ( A ∧ B ) h α i ( A ∨ B ) ⊣⊢ h α i A ∨ h α i B A ⊢ A { α } A ⊢ h α i AA ⊢ { α } h α i A B ⊢ B { α } B ⊢ h α i BB ⊢ { α } h α i BA ∨ B ⊢ { α } h α i A ; { α } h α i BA ∨ B ⊢ { α } ( h α i A ; h α i B ) { α } A ∨ B ⊢ h α i A ; h α i B h α i ( A ∨ B ) ⊢ h α i A ; h α i B h α i ( A ∨ B ) ⊢ h α i A ∨ h α i B A ⊢ AA ⊢ A ; BA ⊢ A ∨ B { α } A ⊢ h α i ( A ∨ B ) h α i A ⊢ h α i ( A ∨ B ) B ⊢ BB ⊢ A ; BB ⊢ A ∨ B { α } B ⊢ h α i ( A ∨ B ) h α i B ⊢ h α i ( A ∨ B ) h α i A ∨ h α i B ⊢ h α i ( A ∨ B ) ; h α i ( A ∨ B ) h α i A ∨ h α i B ⊢ h α i ( A ∨ B ) [ α ]( A ∨ B ) ⊣⊢ α → ( h α i A ∨ h α i B ) A ⊢ A { α } A ⊢ h α i AA ⊢ { α } h α i A B ⊢ B { α } B ⊢ h α i BB ⊢ { α } h α i BA ∨ B ⊢ { α } h α i A ; { α } h α i BA ∨ B ⊢ { α } ( h α i A ; h α i B )[ α ]( A ∨ B ) ⊢ { α } { α } ( h α i A ∨ h α i B ) comp [ α ]( A ∨ B ) ⊢ Φ α > ( h α i A ∨ h α i B )Φ α ; [ α ]( A ∨ B ) ⊢ h α i A ∨ h α i B Φ α ⊢ h α i A ∨ h α i B < [ α ]( A ∨ B )1 α ⊢ h α i A ∨ h α i B < [ α ]( A ∨ B )1 α ; [ α ]( A ∨ B ) ⊢ h α i A ∨ h α i B [ α ]( A ∨ B ) ⊢ α > ( h α i A ∨ h α i B )[ α ]( A ∨ B ) ⊢ α → ( h α i A ∨ h α i B ) Φ α ⊢ α A ⊢ A { α } A ⊢ { α } A h α i A ⊢ { α } A B ⊢ B { α } B ⊢ { α } B h α i B ⊢ { α } B h α i A ∨ h α i B ⊢ { α } A ; { α } B h α i A ∨ h α i B ⊢ { α } ( A ; B ) { α } ( h α i A ∨ h α i B ) ⊢ A ; B { α } ( h α i A ∨ h α i B ) ⊢ A ∨ B h α i A ∨ h α i B ⊢ { α } ( A ∨ B )1 α → ( h α i A ∨ h α i B ) ⊢ Φ α > { α } ( A ∨ B ) reduce ′ α → ( h α i A ∨ h α i B ) ⊢ { α } ( A ∨ B )1 α → ( h α i A ∨ h α i B ) ⊢ [ α ]( A ∨ B ) α i ( A → B ) ⊣⊢ α ∧ ( h α i A → h α i B ) Φ α ⊢ α A ⊢ A { α } A ⊢ { α } A h α i A ⊢ { α } A { α } h α i A ⊢ A B ⊢ B { α } B ⊢ h α i BB ⊢ { α } h α i BA → B ⊢ { α } h α i A > { α } h α i BA → B ⊢ { α } ( h α i A > h α i B ) { α } ( A → B ) ⊢ h α i A > h α i B { α } ( A → B ) ⊢ h α i A → h α i B Φ α ; { α } ( A → B ) ⊢ α ∧ ( h α i A → h α i B ) reduce ′ { α } ( A → B ) ⊢ α ∧ ( h α i A → h α i B ) h α i ( A → B ) ⊢ α ∧ ( h α i A → h α i B ) A ⊢ A { α } A ⊢ h α i A B ⊢ B { α } B ⊢ { α } B h α i B ⊢ { α } B h α i A → h α i B ⊢ { α } A > { α } B h α i A → h α i B ⊢ { α } ( A > B ) { α } ( h α i A → h α i B ) ⊢ A > B { α } ( h α i A → h α i B ) ⊢ A → B { α } { α } ( h α i A → h α i B ) ⊢ h α i ( A → B ) comp αL Φ α ; ( h α i A → h α i B ) ⊢ h α i ( A → B )( h α i A → h α i B ) ; Φ α ⊢ h α i ( A → B )Φ α ⊢ ( h α i A → h α i B ) > h α i ( A → B )1 α ⊢ ( h α i A → h α i B ) > h α i ( A → B )( h α i A → h α i B ) ; 1 α ⊢ h α i ( A → B )1 α ; ( h α i A → h α i B ) ⊢ h α i ( A → B )1 α ∧ ( h α i A → h α i B ) ⊢ h α i ( A → B ) [ α ]( A → B ) ⊣⊢ h α i A → h α i B A ⊢ A { α } A ⊢ { α } A { α } { α } A ⊢ A B ⊢ B { α } B ⊢ h α i BB ⊢ { α } h α i BA → B ⊢ { α } { α } A > { α } h α i BA → B ⊢ { α } ( { α } A > h α i B )[ α ]( A → B ) ⊢ { α } { α } ( { α } A > h α i B ) comp [ α ]( A → B ) ⊢ Φ α > ( { α } A > h α i B )Φ α ; [ α ]( A → B ) ⊢ { α } A > h α i B { α } A ; (Φ α ; [ α ]( A → B )) ⊢ h α i B ( { α } A ; Φ α ) ; [ α ]( A → B ) ⊢ h α i B [ α ]( A → B ) ; ( { α } A ; Φ α ) ⊢ h α i B { α } A ; Φ α ⊢ [ α ]( A → B ) > h α i B Φ α ; { α } A ⊢ [ α ]( A → B ) > h α i B reduce ′ { α } A ⊢ [ α ]( A → B ) > h α i B h α i A ⊢ [ α ]( A → B ) > h α i B [ α ]( A → B ) ; h α i A ⊢ h α i B h α i A ; [ α ]( A → B ) ⊢ h α i B [ α ]( A → B ) ⊢ h α i A > h α i B [ α ]( A → B ) ⊢ h α i A → h α i B A ⊢ A { α } A ⊢ h α i A B ⊢ B { α } B ⊢ { α } B h α i B ⊢ { α } B h α i A → h α i B ⊢ { α } A > { α } B h α i A → h α i B ⊢ { α } ( A > B ) { α } ( h α i A → h α i B ) ⊢ A > B { α } ( h α i A → h α i B ) ⊢ A → B h α i A → h α i B ⊢ { α } ( A → B ) h α i A → h α i B ⊢ [ α ]( A → B ) β ,such that α a β form the set { β i | ≤ i ≤ n } . h α ih a i A ⊢ α ∧ W {h a ih β i A | α a β } Φ α ⊢ α A ⊢ A { β } A ⊢ h β i A { a }{ β } A ⊢ h a ih β i A · · ·· · ·· · · A ⊢ A { β n } A ⊢ h β n i A { a }{ β n } A ⊢ h a ih β n i A swap-out ′ { α }{ a } A ⊢ ; (cid:16) h a ih β i i A (cid:17) { α }{ a } A ⊢ W (cid:16) h a ih β i i A (cid:17) Φ α ; { α }{ a } A ⊢ α ∧ W (cid:16) h a ih β i i A (cid:17) reduce ′ { α }{ a } A ⊢ α ∧ W (cid:16) h a ih β i i A (cid:17) { a } A ⊢ { α } α ∧ W (cid:16) h a ih β i i A (cid:17) h a i A ⊢ { α } α ∧ W (cid:16) h a ih β i i A (cid:17) { α }h a i A ⊢ α ∧ W (cid:16) h a ih β i i A (cid:17) h α ih a i A ⊢ α ∧ W (cid:16) h a ih β i i A (cid:17) α ∧ W {h a ih β i A | α a β } ⊢ h α ih a i A A ⊢ A { a } A ⊢ h a i A { α }{ a } A ⊢ h α ih a i A swap-in ′ Φ α ; { a }{ β } A ⊢ h α ih a i A { a }{ β } A ; Φ α ⊢ h α ih a i A Φ α ⊢ { a }{ β } A > h α ih a i A α ⊢ { a }{ β } A > h α ih a i A { a }{ β } A ; 1 α ⊢ h α ih a i A α ; { a }{ β } A ⊢ h α ih a i A { a }{ β } A ⊢ α > h α ih a i A { β } A ⊢ { a } (1 α > h α ih a i A ) h β i A ⊢ { a } (1 α > h α ih a i A ) { a }h β i A ⊢ α > h α ih a i A h a ih β i A ⊢ α > h α ih a i A · · ·· · ·· · ·· · ·· · ·· · ·· · ·· · ·· · ·· · ·· · ·· · ·· · ·· · · A ⊢ A { a } A ⊢ h a i A { α }{ a } A ⊢ h α ih a i A swap-in ′ Φ α ; { a }{ β n } A ⊢ h α ih a i A { a }{ β n } A ; Φ α ⊢ h α ih a i A Φ α ⊢ { a }{ β n } A > h α ih a i A α ⊢ { a }{ β n } A > h α ih a i A { a }{ β n } A ; 1 α ⊢ h α ih a i A α ; { a }{ β n } A ⊢ h α ih a i A { a }{ β n } A ⊢ α > h α ih a i A { β n } A ⊢ { a } (1 α > h α ih a i A ) h β n i A ⊢ { a } (1 α > h α ih a i A ) { a }h β n i A ⊢ α > h α ih a i A h a ih β n i A ⊢ α > h α ih a i A W (cid:16) h a ih β i i A (cid:17) ⊢ ; (cid:16) α > h α ih a i A (cid:17)W (cid:16) h a ih β i i A (cid:17) ⊢ α > h α ih a i A )1 α ; W (cid:16) h a ih β i i A (cid:17) ⊢ h α ih a i A α ∧ W (cid:16) h a ih β i i A (cid:17) ⊢ h α ih a i A α ] h a i A ⊢ P re ( α ) → W {h a ih β i A | α a β } A ⊢ A { β } A ⊢ h β i A { a }{ β } A ⊢ h a ih β i A · · ·· · ·· · · A ⊢ A { β n } A ⊢ h β n i A { a }{ β n } A ⊢ h a ih β n i A swap-out ′ { α }{ a } A ⊢ ; (cid:16) h a ih β i i A (cid:17) { α }{ a } A ⊢ W (cid:16) h a ih β i i A (cid:17) { a } A ⊢ { α } W (cid:16) h a ih β i i A (cid:17) h a i A ⊢ { α } W (cid:16) h a ih β i i A (cid:17) [ α ] h a i A ⊢ { α } { α } W (cid:16) h a ih β i i A (cid:17) comp [ α ] h a i A ⊢ Φ α > W (cid:16) h a ih β i i A (cid:17) Φ α ; [ α ] h a i A ⊢ W (cid:16) h a ih β i i A (cid:17) Φ α ⊢ W (cid:16) h a ih β i i A (cid:17) < [ α ] h a i A α ⊢ W (cid:16) h a ih β i i A (cid:17) < [ α ] h a i A α ; [ α ] h a i A ⊢ W (cid:16) h a ih β i i A (cid:17) [ α ] h a i A ⊢ α > W (cid:16) h a ih β i i A (cid:17) [ α ] h a i A ⊢ α → W (cid:16) h a ih β i i A (cid:17) P re ( α ) → W {h a ih β i i A | α a β } ⊢ [ α ] h a i A Φ α ⊢ α A ⊢ A { a } A ⊢ h a i A { α }{ a } A ⊢ { α }h a i A swap-in ′ Φ α ; { a }{ β } A ⊢ { α }h a i A { a }{ β } A ⊢ Φ α > { α }h a i A reduce ′ { a }{ β } A ⊢ { α }h a i A { β } A ⊢ { a } { α }h a i A h β i A ⊢ { a } { α }h a i A { a }h β i A ⊢ { α }h a i A h a ih β i A ⊢ { α }h a i A · · ·· · ·· · ·· · ·· · ·· · ·· · ·· · ·· · ·· · · A ⊢ A { a } A ⊢ h a i A { α }{ a } A ⊢ { α }h a i A swap-in ′ Φ α ; { a }{ β n } A ⊢ { α }h a i A { a }{ β n } A ⊢ Φ α > { α }h a i A reduce ′ { a }{ β n } A ⊢ { α }h a i A { β n } A ⊢ { a } { α }h a i A h β n i A ⊢ { a } { α }h a i A { a }h β n i A ⊢ { α }h a i A h a ih β n i A ⊢ { α }h a i A W (cid:16) h a ih β i i A (cid:17) ⊢ ; (cid:16) { α }h a i A (cid:17)W (cid:16) h a ih β i i A (cid:17) ⊢ { α }h a i A α → W (cid:16) h a ih β i i A (cid:17) ⊢ Φ α > { α }h a i A reduce ′ α → W (cid:16) h a ih β i i A (cid:17) ⊢ { α }h a i A α → W (cid:16) h a ih β i i A (cid:17) ⊢ [ α ] h a i A α ][ a ] A ⊢ P re ( α ) → V { [ a ][ β ] A | α a β } A ⊢ A [ a ] A ⊢ { a } A [ α ][ a ] A ⊢ { α }{ a } A swap - in ′ [ α ][ a ] A ⊢ Φ α > { a }{ β } A Φ α ; [ α ][ a ] A ⊢ { a }{ β } A { a } Φ α ; [ α ][ a ] A ) ⊢ { β } A { a } Φ α ; [ α ][ a ] A ) ⊢ [ β ] A Φ α ; [ α ][ a ] A ⊢ { a } [ β ] A Φ α ; [ α ][ a ] A ⊢ [ a ][ β ] A · · ·· · ·· · ·· · ·· · ·· · ·· · ·· · ·· · · A ⊢ A [ a ] A ⊢ { a } A [ α ][ a ] A ⊢ { α }{ a } A swap - in ′ [ α ][ a ] A ⊢ Φ α > { a }{ β n } A Φ α ; [ α ][ a ] A ⊢ { a }{ β n } A { a } (Φ α ; [ α ][ a ] A ) ⊢ { β n } A { a } (Φ α ; [ α ][ a ] A ) ⊢ [ β n ] A Φ α ; [ α ][ a ] A ⊢ { a } [ β n ] A Φ α ; [ α ][ a ] A ⊢ [ a ][ β n ] A ; (cid:16) Φ α ; [ α ][ a ] A (cid:17) ⊢ V (cid:16) [ a ][ β i ] A (cid:17) Φ α ; [ α ][ a ] A ⊢ V (cid:16) [ a ][ β i ] A (cid:17) [ α ][ a ] A ; Φ α ⊢ V (cid:16) [ a ][ β i ] A (cid:17) Φ α ⊢ [ α ][ a ] A >
V (cid:16) [ a ][ β i ] A (cid:17) α ⊢ [ α ][ a ] A >
V (cid:16) [ a ][ β i ] A (cid:17) [ α ][ a ] A ; 1 α ⊢ V (cid:16) [ a ][ β i ] A (cid:17) α ; [ α ][ a ] A ⊢ V (cid:16) [ a ][ β i ] A (cid:17) [ α ][ a ] A ⊢ α > V (cid:16) [ a ][ β i ] A (cid:17) [ α ][ a ] A ⊢ α → V (cid:16) [ a ][ β i ] A (cid:17) P re ( α ) → V { [ a ][ β ] A | α a β } ⊢ [ α ][ a ] A Φ α ⊢ α A ⊢ A [ β ] A ⊢ { β } A [ a ][ β ] A ⊢ { a }{ β } A · · ·· · ·· · · A ⊢ A [ β n ] A ⊢ { β n } A [ a ][ β n ] A ⊢ { a }{ β n } A swap - out ′ ; (cid:16) [ a ][ β i ] A (cid:17) ⊢ { α }{ a } A V (cid:16) [ a ][ β i ] A (cid:17) ⊢ { α }{ a } A α → V (cid:16) [ a ][ β i ] A (cid:17) ⊢ Φ α > { α }{ a } A reduce ′ α → V (cid:16) [ a ][ β i ] A (cid:17) ⊢ { α }{ a } A { α } (1 α → V (cid:16) [ a ][ β i ] A (cid:17) ) ⊢ { a } A { α } (1 α → V (cid:16) [ a ][ β i ] A (cid:17) ) ⊢ [ a ] A α → V (cid:16) [ a ][ β i ] A (cid:17) ⊢ { α } [ a ] A α → V (cid:16) [ a ][ β i ] A (cid:17) ⊢ [ α ][ a ] A α i [ a ] A ⊢ P re ( α ) ∧ V { [ a ][ β ] A | α a β } Φ α ⊢ α A ⊢ A [ a ] A ⊢ { a } Abalance { α } [ a ] A ⊢ { α }{ a } Aswap - in ′ { α } [ a ] A ⊢ Φ α > { a }{ β } A Φ α ; { α } [ a ] A ⊢ { a }{ β } Areduce ′ { α } [ a ] A ⊢ { a }{ β } A { α } [ a ] A ⊢ { a }{ β } A { a } { α } [ a ] A ⊢ { β } A { a } { α } [ a ] A ⊢ [ β ] A { α } [ a ] A ⊢ { a } [ β ] A { α } [ a ] A ⊢ [ a ][ β ] A · · ·· · ·· · ·· · ·· · ·· · ·· · ·· · ·· · ·· · ·· · · A ⊢ A [ a ] A ⊢ { a } A balance { α } [ a ] A ⊢ { α }{ a } A swap - in ′ { α } [ a ] A ⊢ Φ α > { a }{ β n } A Φ α ; { α } [ a ] A ⊢ { a }{ β n } A reduce ′ { α } [ a ] A ⊢ { a }{ β n } A { α } [ a ] A ⊢ { a }{ β n } A { a } { α } [ a ] A ⊢ { β n } A { a } { α } [ a ] A ⊢ [ β n ] A { α } [ a ] A ⊢ { a } [ β n ] A { α } [ a ] A ⊢ [ a ][ β n ] A ; (cid:16) { α } [ a ] A (cid:17) ⊢ V (cid:16) [ a ][ β i ] A (cid:17) { α } [ a ] A ⊢ V (cid:16) [ a ][ β i ] A (cid:17) Φ α ; { α } [ a ] A ⊢ α ∧ V (cid:16) [ a ][ β i ] A (cid:17) reduce ′ { α } [ a ] A ⊢ α ∧ V (cid:16) [ a ][ β i ] A (cid:17) h α i [ a ] A ⊢ α ∧ V (cid:16) [ a ][ β i ] A (cid:17) P re ( α ) ∧ V { [ a ][ β ] A | α a β } ⊢ h α i [ a ] A A ⊢ A [ β ] A ⊢ { β } A [ a ][ β ] A ⊢ { a }{ β } A · · ·· · ·· · · A ⊢ A [ β n ] A ⊢ { β n } A [ a ][ β n ] A ⊢ { a }{ β n } A swap - out ′ ; (cid:16) [ a ][ β i ] A (cid:17) ⊢ { α }{ a } A V (cid:16) [ a ][ β i ] A (cid:17) ⊢ { α }{ a } A { α } V (cid:16) [ a ][ β i ] A (cid:17) ⊢ { a } A { α } V (cid:16) [ a ][ β i ] A (cid:17) ⊢ [ a ] A { α } { α } V (cid:16) [ a ][ β i ] A (cid:17) ⊢ h α i [ a ] A Φ α ; V (cid:16) [ a ][ β i ] A (cid:17) ⊢ h α i [ a ] A V (cid:16) [ a ][ β i ] A (cid:17) ; Φ α ⊢ h α i [ a ] A Φ α ⊢ V (cid:16) [ a ][ β i ] A (cid:17) > h α i [ a ] A α ⊢ V (cid:16) [ a ][ β i ] A (cid:17) > h α i [ a ] A V (cid:16) [ a ][ β i ] A (cid:17) ; 1 α ⊢ h α i [ a ] A α ; V (cid:16) [ a ][ β i ] A (cid:17) ⊢ h α i [ a ] A α ∧ V (cid:16) [ a ][ β i ] A (cid:17) ⊢ h α i [ a ] A eferences [1] Samson Abramsky and Steven Vickers. Quantales, observational logic andprocess semantics. Mathematical Structures in Computer Science , 3:161–227, 6 1993.[2] Peter Aczel.
Non-Well-Founded Sets . CSLI, Stanford, 1988.[3] Peter Aczel and Nax Paul Mendler. A final coalgebra theorem. In
CategoryTheory and Computer Science , pages 357–365, 1989.[4] Jirí Adámek, Stefan Milius, and Jiri Velebil. On coalgebra based on classes.
Theoretical Computer ScienceComput. Sci. , 316(1):3–23, 2004.[5] Jirí Adámek, Stefan Milius, and Jiri Velebil. A general final coalgebratheorem.
Mathematical Structures in Computer Science , 15(3):409–432,2005.[6] Guillaume Aucher. Characterizing updates in dynamic epistemic logic. In
Knowledge Representation and Reasoning , 2010.[7] Guillaume Aucher. DEL-sequents for progression.
Journal of Applied Non-Classical Logics , 21(3-4):289–321, 2011.[8] Guillaume Aucher, Bastien Maubert, and François Schwarzentruber. Gen-eralized del-sequents. In
Logics in Artificial Intelligence , volume 7519 of
Lecture Notes in Computer Science , pages 54–66. Springer, 2012.[9] Philippe Balbiani, Hans van Ditmarsch, Andreas Herzig, and Tiagode Lima. Tableaux for public announcement logic.
Journal of Logic andComputation , 20(1):55–76, 2010.[10] Alexandru Baltag. A coalgebraic semantics for epistemic programs.
Electr.Notes Theor. Comput. Sci. , 82(1):17–38, 2003.[11] Alexandru Baltag, Bob Coecke, and Mehrnoosh Sadrzadeh. Epistemic ac-tions as resources.
J. Log. Comput. , 17(3):555–585, 2007.[12] Alexandru Baltag, Lawrence S. Moss, and Slawomir Solecki. The logic ofpublic announcements, common knowledge and private suspicious. Tech-nical Report SEN-R9922, CWI, Amsterdam, 1999.[13] Michael Barr. Terminal coalgebras in well-founded set theory.
TheoreticalComputer Science , 114(2):299–315, 1993. See also Theoretical ComputerScience 124(1):189-192, 1994.[14] Giulia Battilotti, Claudia Faggian, and Giovanni Sambin. Basic logic: Re-flection, symmetry, visibility.
Journal of Symbolic Logic , 65, 2000.[15] Nuel Belnap. Display logic.
Journal of Philosophical Logic , 11:375–417,1982. 5916] Nuel Belnap. Linear logic displayed.
Notre Dame Journal of Formal Logic ,31(1):14–25, 1990.[17] Corina Cîrstea and Mehrnoosh Sadrzadeh. Coalgebraic epistemic updatewithout change of model. In
CALCO’07 , pages 158–172, 2007.[18] Haskell B. Curry.
Foundations of Mathematical Logic . McGraw-Hill, NewYork, 1963.[19] Roy Dyckhoff, Mehrnoosh Sadrzadeh, and Julien Truffaut. Algebra, prooftheory and applications for an intuitionistic logic of propositions, actionsand adjoint modal operators.
ACM Transactions on Computational Logic ,14(4), 2013.[20] Sabine Frittella and Giuseppe Greco. Display-type sequent calculus formonotone modal logic.
Advances in Modal Logic (2014), Groningen. Shortpresentation , 2014.[21] Sabine Frittella, Giuseppe Greco, Alexander Kurz, and Alessandra Palmi-giano. A multi-type display calculus for game logic.
In preparation .[22] Sabine Frittella, Giuseppe Greco, Alexander Kurz, and Alessandra Palmi-giano. Multi-type display calculus for propositional dynamic logic.
Journalof Logic and Computation - Special Issue on Substructural Logic and In-formation Dynamics. Forthcoming. , 2014.[23] Sabine Frittella, Giuseppe Greco, Alexander Kurz, Alessandra Palmigiano,and Vlasta Sikimić. A multi-type display calculus for dynamic epistemiclogic.
Journal of Logic and Computation - Special Issue on SubstructuralLogic and Information Dynamics. Forthcoming. , 2014.[24] Sabine Frittella, Giuseppe Greco, Alexander Kurz, Alessandra Palmigiano,and Vlasta Sikimić. Multi-type sequent calculi. In Michal Zawidzki An-drzej Indrzejczak, Janusz Kaczmarek, editor,
Trends in Logic XIII , pages81–93. Lodź University Press, 2014.[25] Gerhard Gentzen.
The collected papers of Gerhard Gentzen / edited by M.E. Szabo . North-Holland Pub. Co Amsterdam, 1969.[26] Jelle Gerbrandy.
Bisimulations on Planet Kripke . PhD thesis, ILLC, Uni-versity of Amsterdam, 1999.[27] Jelle Gerbrandy and Willem Groeneveld. Reasoning about informationchange.
Journal of Logic, Language and Information , 6(2):147–169, 1997.[28] Rajeev Goré. Substructural logics on display.
Logic Journal of IGPL ,6(3):451–504, 1998.[29] Rajeev Goré. Dual intuitionistic logic revisited. In Roy Dyckhoff, editor,
Automated Reasoning with Analytic Tableaux and Related Methods , volume1847 of
Lecture Notes in Computer Science , pages 252–267. Springer, 2000.6030] Giuseppe Greco, Alexander Kurz, and Alessandra Palmigiano. Dynamicepistemic logic displayed. In Huaxin Huang, Davide Grossi, and OlivierRoy, editors,
Proceedings of the 4th International Workshop on Logic, Ra-tionality and Interaction (LORI-4) , volume 8196 of
LNCS , 2013.[31] David Harel, Dexter Kozen, and Jerzy Tiuryn.
Dynamic Logic . MIT Press,Cambridge, MA, 2000.[32] Marcus Kracht. Power and weakness of the modal display calculus. In
Proof theory of modal logic , pages 93–121. Kluwer, 1996.[33] Alexander Kurz. Modal rules are co-implications. In
CMCS , volume 44 of
Electronic Notes in Theoretical Computer Science , pages 241–253, 2001.[34] Alexander Kurz. Logics admitting final semantics. In
FoSSaCS , volume2303 of
Lecture Notes in Computer Science , pages 238–249, 2002.[35] Alexander Kurz and Alessandra Palmigiano. Epistemic updates on algeb-ras.
Logical Methods in ComputerScience , 2013. arXiv:1307.0417.[36] Alexander Kurz and Alessandra Palmigiano. Epistemic updates on algeb-ras.
Logical Methods in Computer Science , 2013. abs/1307.0417.[37] Alexander Kurz and Jirí Rosický. Operations and equations for coalgebras.
Mathematical Structures in Computer Science , 15(1):149–166, 2005.[38] Carsten Lutz. Complexity and succinctness of public announcement lo-gic.
AAMAS ’06 Proceedings or the fifth international joint conference onAutonomous agents and multiagent system , pages 137–143, 2006.[39] Minghui Ma, Alessandra Palmigiano, and Mehrnoosh Sadrzadeh. Algebraicsemantics and model completeness for intuitionistic public announcementlogic.
Annals of Pure and Applied Logic , 2013. to appear.[40] Michael Moortgat. Multimodal linguistic inference.
Logic Journal of theIGPL , 3(2-3):371–401, 1995.[41] Sara Negri and Paolo Maffezioli. A proof theoretical perspective on publicannouncement logic.
Logic and Philosophy of Science , 9:49–59, 2011.[42] Sara Negri and Paolo Maffezzoli. A Gentzen-style analysis of public an-nouncement logic. pages 293–313. University of the Basque Country Press,2010.[43] Rohit Parikh. The logic of games and its applications. In
Annals of DiscreteMathematics , pages 111–140. Elsevier, 1985.[44] Jan Plaza. Logics of public communications.
Synthese , 158(2):165–179,2007. 6145] Greg Restall.
An Introduction to Substructural Logics . Routledge, London,2000.[46] Peter Schroeder-Heister. Validity concepts in proof-theoretic semantics. InR. Kale and Peter Schroeder-Heister, editors,
Proof-Theoretic Semantics ,volume 148, pages 525–571. 2006.[47] Peter Schroeder-Heister. Proof-theoretic semantics. In Edward N. Zalta,editor,
The Stanford Encyclopedia of Philosophy . Spring 2013 edition, 2013.[48] Hans P. van Ditmarsch, Wiebe van der Hoek, and Barteld Kooi.
DynamicEpistemic Logic . Springer, 2007.[49] Heinrich Wansing.
Displaying Modal Logic . Kluwer, Dordrecht, 1998.[50] Heinrich Wansing. The idea of a proof-theoretic semantics and the meaningof the logical operations.
Studia Logica , 64(1):3–20, 2000.[51] Heinrich Wansing. Sequent systems for modal logics.