Proof Search for Modal Logical systems (S4, S5, GL) Ramyaa Date of your defense Department of Philosophy Carnegie Mellon University Pittsburgh, PA 15213 Thesis Committee: Wilfried Sieg Joseph D. Ramsey A Thesis Submitted for the Degree of Master of Science in logic, Computation and Methodology Copyright 2012 Ramyaa 1 Contents 1. Background 1 1.1. Modal Logic 2 1.2. Logic of Provability 4 1.3. Natural deduction 12 2. Proof search calculi 23 2.1 Systems S4, S5 and GL 23 2.2 Natural Deduction Rules 30 2.3 Intercalation Calculus Rules 34 2.4 Soundness 35 2.5 Completeness 39 3. Implementation 51 3.1 AProS 51 3.2 Proof search in Modal Logic 54 3.3 Implementational Details 56 3.4 Examples 62 Appendix 65 References 117 2 CHAPTER 1 Background This thesis presents the work done on proof search for the modal logical systems S4, S5 and the Logic of Provability (GL). An intercalation calculus ([10]) was used as the underlying logical calculus, and the proof search was automated using the theorem prover AProS [1]. The inference rules in the intercalation calculus for the systems S5 and GL, and their soundness and completeness results were based on the work done for the system S4 in [9]. This chapter gives the theoretical background. Chapter 2 describes the systems GL, S4 and S5, intercalation calculus rules along with their soundness and completeness results. This gives the framework for proof search in these systems. Chapter 3 explains the implementation of the automated proof search procedure. The current chapter is organized as follows: section 1 introduces modal logic - language and semantics of (classical propositional) modal logical systems, in particular normal modal logical systems; section 2 discusses the Logic of Provability, - formal systems and provability, provability represented inside the systems, and properties that a logical system reasoning about provability should have, along with a brief sketch of the use of modal logic as the logic of provability; section 3 gives details about natural deduction systems including Prawitz’ system, normal proofs and intercalation calculus. 1.1 Modal Logic An expression used to qualify the truth of a statement is called a modal – for instance, "it is necessary that …”. Modal logic is in part the study of reasoning with concepts that qualify truth – strictly speaking, the notions of "necessity" (and its dual, "possibility"). Modal logic can 3 also be used to reason about a variety of other modal expressions, such as knowledge (and belief), obligation (and permission), and provability (and consistency.) This thesis deals with the modal logic of provability and consistency - the Logic of Provability (GL), and with the modal logics of necessity and possibility – systems S4 and S5. These systems are normal modal logical systems with Kripke's possible worlds semantics. This section gives - language of classical propositional modal logic, - axioms and inference rules of normal modal logical systems - Kripke semantics of modal logical systems. 1.1.1 Language The language of a modal logical system contains sentential letters Pi (i N), logical connectives (&, V, →, ↔, ¬) and the modal operator (□). The operator ◊ is defined as ¬□¬1. The set of sentences of GL, S4 and S5 is the minimum set containing: i. All sentential letters, ii. (φ1 & φ2), (φ1 V φ2), (φ1 → φ2), (φ1 ↔ φ2), (¬φ1) and (□φ1), where φ1 and φ2 are sentences. 1.1.2 Normal modal logical systems A normal modal logical system contains the following axioms: i. all propositional tautologies, ii. distribution axioms i.e., all sentences of the form (□(φ1→φ2) → (□φ1→□φ2)). 1 The connective □ is used to represent necessity, knowledge, obligation or provability and its dual ◊ is used to represent possibility, belief, permission or consistency. 4 Such a system is also closed under the following operations: i. modus ponens, ii. necessitation i.e., if φ is provable, so is □φ. An arbitrary normal modal logical system contains the axioms and inference rules described above, along with some additional axioms that distinguish different such systems from each other. The axioms for the normal systems S4, S5 and GL are given in chapter 2. A logical system S proves a formula φ, written as S ⊦ φ, if there exists a finite sequence of formulae whose last formula is φ, such that each formula in the sequence is either an axiom of S or follows from preceding formulae by one of the inference rules of S. 1.1.1 Semantics Kripke’s possible worlds semantics is used as the semantic model for the systems studied. The Kripke model for a modal logical system is a triplet M = 〈W, R, ⊩〉, where i. W is a non-empty set, ii. R is a binary relation on the elements of W, and iii. ⊩ is a binary relation between elements of W and formulae. The elements of the set W are known as possible worlds. R is called the accessibility relation; for any two elements, u and v of W, if uRv holds, then v is said to be accessible to u. The relation ⊩ is called valuation. For propositional connectives, this relation mimics the truth value assignment of propositional logic. Regarding □, for any element u of W, u⊩□φ if and only if (∀v)(uRv → v⊩φ). 5 In a given model M, a sentence φ is said to be true at a world u (u ∈ W) if and only if u⊩φ. A sentence φ is said to be valid in a model M = 〈W, R, ⊩〉 if and only if for all u in W, φ is true at u. (We write this as M⊩φ). A sentence φ is said to be satisfiable in a model M = 〈W, R, ⊩〉 if and only if for some u in W, φ is true at u. Finally, we can define when a sentence φ is a semantic consequence of a set of sentences Γ and write Γ⊩φ. This relationship holds just in case, for any model M such that M⊩ψ for all ψ Γ, it also holds that M⊩φ. Given a model M, the properties of its accessibility relation, such as reflexivity, symmetry, transitivity etc. determine it as a model for a particular modal logic. The properties of the accessibility relation of a model required to make it a model of S4, S5 and GL are described in chapter 2. Thus all models considered are models of some normal modal logical system. Since all systems considered here are normal modal logical systems, all propositional tautologies and all distribution axioms are valid in them, as are the sentences derived using modus ponens, and necessitation. 1.2 Logic of Provability The Logic of Provability is a modal logic which studies the concept of formal provability (and consistency). The system considered here is the modal logical system called GL after Gödel and Löb. GL was introduced by Boolos [2]. This section introduces formal systems (in particular Peano Arithmetic), provability within a formal system, the incompleteness theorems, conditions to derive them and a brief description of GL (showing that it reasons about provability in PA). 6 1.2.1 Formal systems and provability Peano Arithmetic (PA) is a formal system whose axioms are the axioms of classical firstorder logic (including those for falsum), axioms for zero and successor, recursion axioms for addition and multiplication, and the induction axiom scheme. PA’s inference rules are modus ponens and generalization. A proof of a formula φ in a formal axiomatic system S is a finite sequence of formulae whose last formula is φ, such that each formula is either an axiom of S or follows from preceding formulae by one of the inference rules of S. If there is a proof of the formula φ in S, φ is said to be provable in (or a theorem of) S, written as S ⊦ φ. A formula φ is refutable in S if the negation of φ is provable in S. A formal system is consistent if it does not both prove and refute a sentence i.e., S does not prove a contradiction; S is (syntactically) complete if for every sentence φ, φ is either provable or refutable in S.. Peano Arithmetic reasons about arithmetic. In order to reason about concepts like provability inside PA, these concepts have to be arithmetized. This “arithmetization of metamathematics” (called Gödel numbering is done by mapping syntactic objects such as formulae into a number (called its Gödel number) using a constructive, one-to-one mapping. For this process, it is necessary to have names within the system for syntactic objects; in Peano Arithmetic, when x′ is the successor of x, the number 3 has the name 0‴; the latter syntactic object is called a numeral; if the Gödel numbering would associate 3 to the conjunction symbol “&”, then 0‴ would be the Gödel numeral for “&”. Here, the Gödel numeral of a syntactic object A is written in bold font where A is the numeral for the Gödel number of A. Using this, all syntactic objects (variables, constants, connectives, formulae, proofs, etc.) except theorems are 7 recursive (i.e., given an object O, it is decidable whether or not O is an object of type T, say of type formulae); theorems are recursively enumerable (i.e, given an object O, it is semidecidable whether O is a theorem). This is explained in more detail below. As mentioned, to reason about concepts like provability inside PA, they need to be represented inside PA, i.e., a predicate in PA should, through Gödel numbering, represent the proof relation (⊦) of PA. As shown by Gödel, there is a predicate binary proof in PA such that if x is the Gödel number of a proof for the formula whose Gödel number is y, then PA proves proof(x, y), and if it is not the case that x is a proof of y, then PA proves ¬proof(x, y). Using this, a predicate Bew(y), (provable(y)) can be formulated as (∃x)proof(x, y). Bew(x) is called the provability predicate or theorem predicate in PA. If PA proves φ, then PA proves Bew(φ). This cannot be proved inside PA, i.e., PA cannot prove (φ → Bew(φ)) for all sentences φ2. We cannot prove its converse i.e., we cannot prove in PA (Bew(φ) → φ) for all sentences φ. In reasoning about provability via a logical system, we aim to capture this predicate Bew. Incompleteness theorems If a formal system is incomplete, there exists a sentence that is neither provable nor refutable in the system. This limits the provability inside the system, and thus very relevant to the logic that reasons about the system’s provability, i.e., logic of provability for that system. Let Z be an arbitrary theory which contains a modicum of number theory and can prove basic arithmetical facts, represent provability etc3. Provability of Z is represented inside Z by a predicate Bewz (analogous to the provability predicate Bew in PA). Consistency is represented in PA can prove this for all Σ1 sentences. Although we are interested in provability of PA, the incompleteness theorems hold for more general systems. Hence, we are going to consider such general systems which can represent number theory and prove some basic arithmetic facts. 2 3 8 the system as "cannot prove a contradiction" i.e., ¬Bewz(⊥), where ⊥ is a placeholder for any contradiction. Gödel's First Incompleteness Theorem (Rosser’s version): If Z is consistent, there is a statement φ, which is neither provable nor refutable in Z. Gödel's Second Incompleteness Theorem: If Z is consistent and satisfies the derivability conditions (described below), then the consistency of Z is not provable inside Z ([4]). Related theorems Löb’s Theorem: Given a formula φ for which Z proves the reflection principle (Bewz(φ)→φ), Z also proves φ ([6]), i.e., If Z ⊦ Bewz(φ) → φ4, then Z ⊦ φ. The second incompleteness theorem can be directly derived from Löb’s theorem. Self-Reference-Lemma: Given any formula φ(y) of Z in which y is the only free variable, there exists a sentence ψ of the language of Z such that Z ⊦ ψ ↔ φ(ψ). This lemma is used in crucially proving the first incompleteness theorem and Löb’s theorem. The concept of provability is reflected inside a formal system by the provability predicate Bewz. Hence, a logical system that reasons about provability needs a version of the Bewz predicate which should mirror Bewz in terms of its power. As the incompleteness theorems are relevant to provability inside the system, the version of the Bewz predicate used should have the properties of Bewz that are needed to prove the incompleteness results. These properties are discussed below. 4 Bewz(φ) → φ is the reflection principle and Löb theorem states that this is provable only if φ is. 9 1.2.2 Derivability conditions Hilbert, Bernays and Löb proved that given an arbitrary theory Z (as above), and an arbitrary formula B(x) of Z, the second incompleteness theorem for Z (with B(x) playing the role of Bewz(x)) is derivable in Z if the following conditions hold ([5], [6]). For any two sentences φ and ψ of Z, Dl: if Z ⊦ φ, then Z ⊦ B(φ) (semirepresentability of theorem predicate) D2: Z ⊦ B(φ → ψ) → (B(φ) →B(ψ)) (provable closure under modus ponens) D3: Z ⊦ B(φ) → B(B(φ)) (formalization of semirepresentability) All these conditions hold for the predicate Bew for PA. Hence, the second incompleteness theorem can be proved for PA. In addition to these three conditions, the self-reference lemma is needed to capture provability. For instance, consider Löb’s theorem formalized as DL: Z ⊦ B(B(φ) → φ) → B(φ) A system Z with a predicate B satisfying D1, D2 and D3 uses the self-reference lemma to prove DL. System Z as described can prove the self-reference lemma. So, this is not an issue. However, if a modal logic is used to capture provability (i.e., has a representation of the predicate B, along with axioms that capture the derivability conditions), it is easy to represent D1, D2, and D3 (in terms of the predicate B), but not the self-reference lemma. However, Z with a predicate B satisfying D1, D2 and DL can prove D3 without using the self-reference lemma (also shown below). Further, DL can be represented in terms of B. Due to this, the Logic of Provability is defined to have a predicate B(x) such that D1, D2 and DL are true for it. 10 Equivalence of D3 and DL (in the presence of D1 and D2) (i) From Dl, D2, and D3, Löb’s theorem can be derived (using self-reference-lemma). Proof: Consider the sentence B(x) → φ with one free variable, x. By the Self-Reference-Lemma, there is a sentence ψ such that (1) Z ⊦ ψ↔(B(ψ) → φ) Self-Reference-Lemma (2) Z ⊦ ψ→ (B(ψ) → φ) from (1) (3) Z ⊦ B(ψ→ (B(ψ) → φ)) by D1 (4) Z ⊦ B(ψ→ (B(ψ) → φ)) →(B(ψ)→ B(B(ψ) → φ)) by D2 (5) Z ⊦ B(ψ)→ B(B(ψ) → φ) from (3), (4) (6) Z ⊦ B(B(ψ) → φ) → (B(B(ψ)) →B(φ)) by D2 (7) Z ⊦ B(ψ) → (B(B(ψ)) →B(φ)) from (5), (6) by transitivity (8) Z ⊦ B(ψ) → B(B(ψ)) by D3 (9) Z ⊦ B(ψ) → B(φ) using (7), (8) (10) Z ⊦ (B(φ) → φ) → (B(ψ) → φ) using (9) (11) Z ⊦ (B(φ) → φ) → ψ 10,1 transitivity (12) Z ⊦ B((B(φ) → φ) → ψ) by D1 (13) Z ⊦ B((B(φ) → φ) → ψ) → (B(B(φ) → φ) → B(ψ) by D2 (14) Z ⊦ B(B(φ) → φ) → B(ψ) from 12, 13 Hence, DL: (15) Z ⊦ B(B(φ) → φ) → B(φ) 14, 9 transitivity Note: This proof can be divided into inferences about the provability predicate (which involve the derivability conditions) and those that do not. This distinction can be formalized as metatheory (which reasons about the provability) and object theory. Though we are interested in provability, Z need not differentiate between meta-theory and object-theory, and inference rules 11 that allow to shift between them. So, the proof above does not make this distinction. The proof formalized in the appendix is done in AProS, and makes this distinction explicit. (ii) From Dl, D2 and DL, D3 can be proved. Proof: Distribution of B over conjunction is assumed and not proved here. (1) Z ⊦ (B(B(φ) & φ) → (B(B(φ)) & B(φ)) &distribution5 (2) Z ⊦ (B(B(φ) & φ) → B(φ) from (1) (3) Z ⊦ φ → ((B(B(φ) & φ) → (B(φ) & φ)) from (2) (4) Z ⊦ B(φ→((B(B(φ) & φ) → (B(φ) & φ)) from D1 (5) Z ⊦ B(φ→(B(B(φ) & φ) → (B(φ) & φ)) → (B(φ)→(B(B(B(φ) & φ) → (B(φ) & φ)))) by D2 (6) Z ⊦ B(φ)→(B(B(B(φ) & φ) → (B(φ) & φ))) from (4), (5) (7) Z ⊦ B(B(B(φ) & φ) → (B(φ) & φ)) → B(B(φ) & φ) by DL (8) Z ⊦ B(φ) → B(B(φ) & φ) from (6), (7) (9) Z ⊦ B(φ) → (B(B(φ)) & B(φ)) &distribution Hence Z ⊦ B(φ)→ (B(B(φ)) These theorems were formulated in AProS and the proofs are attached at the end of the chapter. 1.2.4 Logic of provability for PA A logic of provability for PA focuses on the provability predicate. One way to do this is to use a modal logic with □ representing the predicate Bew. (The dual ◊, defined as ¬□¬, represents consistency.) Since □ represents the provability predicate, it has to satisfy the derivability conditions. Any normal modal logical system will satisfy D1 and D2 (D1 correponds to 5 &distribution: B(A&C) → B(A) & B(C). Proof sketch: 1.(A&C) → A; 2.B((A&C) → A) by D1; 3. B((A&C) → A) →(B(A&C) → B(A)) by D2; 4.B(A&C) → B(A) from 2,3. Similarly for C; &-introduction gives then B(A&C) → B(A) & B(C). 12 necessecitation, and D2 to the distributivity axiom). Adding D3 (□A→□□A) gives the system K4. But, K4 cannot prove some theorems about provability, such as Löb’s theorem. As mentioned earlier, this is because K4 cannot prove the self-reference-lemma (or even represent it), which is needed for proving Löb’s theorem. Adding DL (□(□A→A)) instead of D3 gives GL. Since Löb’s theorem can prove D3 without using self-reference lemma, GL can prove D3 i.e., K4 ≤ GL. This can be shown directly by showing that GL proves D3. This proof proceeds exactly as the proof (ii) above. Proof (ii) gives the proof inside a system Z that is similar to PA, and simply replacing the predicate B by □ gives the proof in GL, as all the axioms and inference rules used in proof (ii) are available in GL. The normal modal logical system GL (described in chapter 2) captures the provability of PA. This is proved by translating GL sentences into PA sentences by means of a translation function. A realization (^) is a function that assigns to each sentence letter of GL a sentence of the language of PA. A Translation (*) of the modal sentence under realization ^ is defined inductively as: (1) ⊥* = ⊥ (2) P* = P^ where P is a GL sentence letter (3) (φ # ψ)* = (φ* # ψ*) where # is a binary logical connective and φ, ψ are GL sentences (4) (¬φ)* = ¬ (φ)* where φ is a GL sentences (5) (□φ)* = Bew(φ*) where φ is a GL sentence Thus, the translation of a GL sentence is a sentence of PA. Solovay 's completeness theorem: GL ⊦ φ if and only if for all realizations ^, PA ⊦ φ* 13 ([8]). An arbitrary sentence of GL gets mapped to different sentences of PA under different realizations. However, any sentence of the form □φ gets mapped into a sentence of the form Bew(x) under all realizations. Since GL proves exactly those statements whose translations are provable under all realizations, it reasons about provability inside PA. 1.3 Natural Deduction Natural deduction (ND) calculi formalize logical reasoning via inference rules which correspond to steps used in informal proofs. Thus, when mathematical proofs are formalized in ND systems, their structure can usually be preserved. The following sections describe two classical propositional systems - Prawitz' natural deduction system ([7]) and Sieg's intercalation calculus ([10]). First, Prawitz' system is presented, and then the concept of normal proof is explained, after which the intercalation calculus is introduced which allows to search for normal proofs directly. 1.3.1 Prawitz’ Natural Deduction Calculus The classical propositional logic system considered here uses the logical connectives &, ∨, →, ↔ and ¬6. Inference rules: In a natural deduction calculus, the properties of each connective are expressed using a pair of inference rules - an introduction rule and an elimination rule. Each rule infers one formula called the conclusion from one or more formulae called the premises. An introduction rule for a particular connective, #, is used to infer a formula whose main connective is #. An elimination 6 In Apros, falsum is used only in specifying proofs. In some systems (eg: Prawitz’s system) falsum is taken as a basic atomic formula and negation of A is defined using falsum by A → ⊥. 14 rule for a particular connective, #, uses a formula whose main connective is # as a premise (along with other formulae) to infer a conclusion. These rules are specified in the table below. Inference rules are written with premises above their conclusion with a horizontal line between them as in the table or as 〈φl, φ2, …, φn | ψ 〉 where φi (0 ≤ I ≤ n) are the premises and ψ is the conclusion. The premise φ in the rules 〈φ, (φ→ψ) | ψ〉, 〈φ, (φ↔ψ) | ψ〉, the premise χ in the rule 〈( φ∨ψ), χ, χ | χ 〉, and the premise ψ in the rule 〈ψ, (φ ↔ψ) | φ〉, are called minor premises. A premise that is not a minor premise is called the major premise. Note that some rules include derivations, which are defined and explained below. These inference rules discharge assumptions. This is indicated in the inference rules using "[ ]". &I: &E: φ ψ φ&ψ ∨I: φ&ψ φ&ψ φ ψ ∨E: φ ψ φ∨ψ φ∨ψ [φ] [ψ] D1 φvψ χ φ∨ψ χ χ D2 χ χ D1(2) a derivation from φ(ψ) to χ →I: [φ] →E: D φ φ→ψ ψ ψ φ→ψ ψ D a derivation from φ to ψ 15 ↔I: [φ] [ψ] D1 D2 ψ φ ↔E: ψφ↔ψ φ φ φ↔ψ ψ φ↔ψ ψ φ D1 a derivation from φ to ψ D2 a derivation from ψ to φ ¬I: ¬E: [φ] [¬φ] D D ⊥ ⊥ ¬φ φ D a derivation from φ to ⊥ D a derivation from ¬φ to ⊥ ⊥I: φ ¬φ ⊥ φ should be different from ⊥. Formula-trees A rule application obtains a consequence from some premises by means of one of the inference rules, i.e., a particular inference rule is applied to particular formula instances (premises) to infer a particular formula instance (conclusion.) This is indicated by writing the premises and the consequence (below the premises) separated by a horizontal line (as in a rule application). Informally, rule applications can be joined so that the conclusion of one rule application acts as the premise of another rule application to infer a new conclusion. To ensure that the rules 16 are not joined in a circular fashion, the structure built by joining rule applications should be a tree. That will be insured by the inductive generation of formula trees discussed next. Formula-trees are trees whose nodes are formulae. Nodes are distinct from one another, though they may be associated with the same formula. Formula-trees are defined inductively as: i. A formula is a formula tree. ii. If Ф1 Ф2,…, Фn is a sequence of formula trees, then so is Ф1 Ф2, …, Фn ψ where there exists a rule application R, whose premises are the roots of Ф i', and whose conclusion is ψ (the root of the newly generated tree). The edges of the formula tree are the horizontal line in the rule application that separates the premises from the conclusion. The edges of the trees can be annotated with the name of the rule that was applied. Notions concerning trees (to be applied to formula trees): This section defines some terms pertaining to formula trees. The definitions presented here differ in some cases from the ones given by Prawitz in [7]. An occurrence is a formula at a certain place (node) in the formula-tree. Thus the same formula appearing twice at two distinct nodes would correspond to two different occurrences. A path from a formula occurrence φ to a formula occurrence ψ, in a tree is a sequence 〈φ1,…,φn〉 where φ1 is φ, φn, is ψ, and φi is a premise of a rule application whose conclusion is φi+1. A formula occurrence φ is above a formula occurrence ψ in a formula tree (ψ is said to be below φ) if ψ occurs in a path from φ to the goal (root of the formula tree) and ψ is distinct from φ. A top formula (also called assumption) in a formula tree is an occurrence that does not have an occurrence above it i.e., it is a leaf node. The 17 end formula (also called the conclusion) in a formula tree is an occurrence that does not have an occurrence below it, i.e., it is the root of the tree. The height of a formula tree is the number of formula occurrences on the longest path (from any occurrence to any other occurrence). A branch is a path whose first formula is a top formula. The subtree of a formula tree F determined by an occurrence of φ is the tree obtained from F by removing all the occurrences in it except φ and the ones above φ. Proof-trees A proof-tree (or a proof or a deduction or a derivation) for φ is a formula tree whose conclusion is φ and every top formula of the tree is a discharged assumption. If there exists a proof for φ, then φ is said to be provable or to be a theorem (written as ⊦φ). A proof-tree (or a proof) for φ from assumptions Γ is a formula tree such that every top formula is either a discharged assumption or an element of Γ. If there exists a proof of φ from Γ, then φ is said to be provable from Γ (written as Γ⊦φ). To determine whether a formula is provable from some assumptions, a proof search algorithm has to search the space of formula trees to find a proof tree from the assumptions to the conclusion. The space that an algorithm searches is called its search space. The search space consisting of all formula trees is large and unwieldy, but it can be reduced by considering normal proofs, as explained below. Normal proofs A proof is said to be normal if it contains no formula occurrence that is both the conclusion of an application of an I-rule (or the falsum rule) and the major premise of an application of an E-rule. Prawitz established that in a restricted version of classical logic7 a proof 7 the language is restricted to ⊥, &, → and ; and the ¬E-rule allows only atomic formulae as its conclusion. 18 of φ from Γ can be converted into a normal proof of φ from Γ [7]. The full result for the system described above was given by a number of people. (See references in Troelstra and Schwichtenberg[11] as well as Byrnes [3].) In a normal proof, every formula is either a subformula of an open assumption or of the conclusion (or the negation of a formula that has been inferred by ¬E). A proof search algorithm can make use of this fact to restrict its search space. However, in the natural deduction system described, the normal proofs are not inductively specified, i.e., there is no direct way to generate all and only the normal proofs. Intercalation calculi were introduced to specify and allow the search for normal proofs. 1.3.2 Intercalation Calculus The intercalation (IC) calculus ([10]) provides a framework which allows to search directly for normal natural deduction proofs. The object of proof search is to find a normal proof of the sentence G (called the goal) from a sequence of sentences α (called the assumptions.) The proof search in the intercalation calculus is to reflect the informal idea of finding a way to close the gap between the goal and the assumptions via logical rules. IC-rules are divided into categories as follows: (i) Elimination rules which are applied to available assumptions to infer new formulae which can then be added to the available assumptions to infer the current goal. In case of ∨E, the rule adds the one disjunct to the available assumption, then the other, requiring the goal to be proved using either disjunct. (ii) Introduction rules which have as their conclusion the current goal. The premises of these rules generate new goals to be proven from the available assumptions. Rules like →I and ↔I add to the available assumptions, in addition to giving a new goal. (iii) Negation rules – to be used in indirect reasoning. 19 Notations Let capital letters G, H…, and the lower case Greek letters φ (and φi (i ∈ N)) denote individual formulae. Let lower case Greek letters α, β… denote finite sequences of formulae. Let αβ denote the concatenation of the sequences α and β. Let α,G denote the sequence α concatenated with the sequence containing the single formula G. Let φ ∈ α denote that φ is an element of the sequence α. IC-rules IC-rules operate on triplets (called questions) of the form 〈α; β? G〉 where α is the sequence of available assumptions; G is the current goal; β is a sequence of formulae obtained by &-elimination and →-elimination from elements in α. The question 〈α?G〉 denotes 〈α;{}?G〉. The IC-rules are presented as (question1 ⇒ questioni (i ∈ N)). An IC-rule is said to be applied to question1 generates a new question(s) – questioni. As mentioned, they are categorized into elimination, introduction and negation rules. Elimination rules: &-E: α; β ? G, (φ1&φ2) ∈ αβ, φi∉αβ V -E: α; β ? G, (φ1∨φ2)∈αβ, φ1∉αβ, φ2∉αβ ⇒ α,φ1; β ? G AND α,φ2; β ? G →-E: α; β ? G, (φ1→φ2)∈αβ, φ2∉αβ ⇒ α; β ? φ1 AND α; β, φ2 ? G ↔-E: α; β ? G, (φ1↔φ2)∈αβ, φi∉αβ ⇒ α; β,φi ? G AND α; β? φ3-i (i = 1 OR 2 ) ⇒ 20 α; β,φi ? G, ( i = 1 OR 2) Introduction rules: &-I: α; β ? (φ1&φ2) ⇒ α; β ? φ1 AND α; β ? φ2 -I: α; β ? (φ1 V φ2) ⇒ α; β ? φ1 OR α; β ? φ2 →-I: α; β ? (φ1→φ2) ⇒ α, φ1; ? φ2 ↔-I: α; β ? (φ1↔φ2) ⇒ α, φ1; ? φ2 AND α, φ2; ? φ V 1 Negation Rules: ¬E: α; β ? φ, φ ≠⊥ ⇒ α, ¬φ; ? ⊥ ¬I : α; β ? ¬φ, ⇒ α, φ; ? ⊥ ⊥I : α; β ? ⊥, ¬φ ∈ F(α) ⇒ α; β ? φ AND α; β ? ¬φ where F(α) is the class of all strictly positive subformulae of elements in α. IC-tree Informally, we can join IC-rule application similar to rule application in ND i.e., IC-rule R1 can act on the question generated by the IC-rule R2. Rule applications are joined to form ICtrees. Though this is similar to the notion of a formula tree, there are differences as explained below. An IC-tree is a tree whose nodes are questions (i.e., they contain available assumptions and the goal to be proven) and whose edges are the IC- rules connecting them. At each node, all possible rule applications are used to extend the tree. IC-tree for the 〈α; β ? G〉 is specified inductively as the tree generated by applying (all applicable) IC-rules to it or to the leaves of an already obtained partial tree (leaves have to be “non-terminal” as explained in the algorithm below). 21 The algorithm to construct an IC-tree for 〈α ? G〉 works as follows: i. Create a tree with 〈α ? G〉 as its root. ii. At each leaf node 〈α; β ? G〉, the rules are applied as follows: (a) If G∈αβ, then the branch8 containing the node is closed with a Y (terminal node). (b) If G∉αβ, and every applicable rule leads to a node that is equivalent9 to one that is already present in the branch, the branch containing the node is closed with an N (terminal node). (c) If G∉αβ, and there is an applicable rule that leads to a node that is not equivalent to any that is already present in the branch, the tree is extended at the node by applying all applicable rules and adding all questions generated (non-terminal nodes). The algorithm terminates in step (ii), when there are no leaf nodes that are questions i.e., all the branches are closed with terminal nodes. There are finitely many IC-rules, and finitely many formulae to which they can be applied. Further, any new formula contained in a newly generated question is a subformula of an existing question. Since repetitions are not allowed, the IC-tree is finite and the algorithm always terminates, generating the IC-tree. IC-proof An IC-proof of G from α is a subtree T of the IC-tree Σ for 〈α; ? G〉 satisfying: i. 〈α; ? G〉 is the root of T, 8 Branch is defined as in formula trees – as a sequence of joined rule applications from a leaf node. A node 〈α; β ? G〉 is equivalent to the node 〈α′; β′ ? G〉 if the set of formulae present in the sequences αβ and α′β′ are identical. 9 22 ii. all the branches of T are Y-closed branches of Σ, iii. every question node (node corresponding to a question) in T that is not the root is followed by exactly one rule. The definition can be extended for and IC-proof of G from α;β in the obvious way. IC-rules for classical propositional logic can be proved to be sound and complete. The completeness result states that the IC-tree for G from assumptions α contains either an IC-proof, or a branch from which a counterexample to the inference of α from G can be constructed [(9]). It can be proved inductively on the height of IC-proofs that any IC-proof of G from assumptions in α can be transformed into a normal natural deduction proof of G from the same assumptions. Thus, if an IC-proof exists for G from assumptions in α, then a normal natural deduction proof exists. Using this proof extraction theorem and the completeness result, a sharpened completeness result can be given stating that the IC-tree for the question 〈α; ? G〉 allows us to determine either a normal proof of G from α or to construct a counterexample to the inference from α to G ([10]). An automated algorithm for efficiently generating an IC-subtree that is an IC-proof is implemented in AProS ([1]). The IC calculus is based on Gentzen's sequent formulation of natural deduction systems. The main differences between Gentzen's sequent formulation and the above formulation are as follows. First, the elimination rules are applied only on the left side of the “?” and introduction rules only on the right side (which ensures normality). Second, negation is taken to be a basic connective, with ⊥ taken to be a placeholder for contradictions (used in indirect arguments). Thus a variety of strategy restrictions can be brought to bear. 23 IC-rules (with soundness and completeness results) for modal logical systems S4, S5 and GL are presented in the following chapter. 24 CHAPTER 2 Proof search calculi This chapter describes the theoretical background for the proof search in the modal logical systems S4, S5 and GL. The rules for these systems were formulated in the intercalation calculus and implemented in the automated theorem prover AproS [1]. The first section introduces the systems, giving the axioms and semantics. The second section gives the natural deduction rules that would allow normal derivations. The third section gives the IC rules, their soundness and completeness proofs. The natural deduction rules were formulated based on the rules for S4 and S5 given by Prawitz [7]. The IC rules and completeness theorems are based on the rules and completeness results given for S4 by Sieg and Cittadini in [9]. 2.1 Systems S4, S5 and GL S4, S5 and GL are classical propositional systems. Their language was presented in chapter 1. 2.1.1 Axioms for S4, S5 and GL: As mentioned earlier, S4, S5 and GL are normal systems. Hence their axioms include (i) Necessitation (from ⊦conclude ⊦□ (ii) distribution axioms (from ⊦□(→conclude ⊦□→□ In addition the system S4 contains (a) □ → (b) □ → □□ In addition to the axioms (i) (ii), (a), and (b), the system S5 contains: (c) □ → □□ 25 In addition to the axioms (i), (ii) of the normal systems, GL contains: (d) □(□ → ) → □ The systems are closed under modus ponens. As mentioned in chapter 1, (i) corresponds to derivability condition D1 and (ii) corresponds to D2, and (d) corresponds to DL. Note that (b) corresponds to D3 and the axiom can be proved in GL. GL, does not have the axiom →□(necessitation i.e., if GL ⊦ then GL ⊦ □ is not internalized) as this would correspond to PA ⊦ φ→Bew(φ), (this is true if φ is a Σ1 sentence.) 2.1.2 Semantics Kripke models for modal logic were presented in chapter 1.1.3. Given a model M, the properties of its accessibility relation R makes it a model for a particular modal logical system. can be derived in a modal logical system L if is true in every world, in every model of L (i.e., a model with the accessibility relation corresponding to L) under every evaluation. Here, we describe the properties of accessibility relations that correspond to axioms of S4, S5 and GL. Axioms true in any normal modal logical system: Necessitation (D1): By the definition of the valuation relation, all tautologies are true in all worlds, (including those accessible from the current one). So, □ is true for all tautologies . Distribution (D2): Assume w⊩□(→andw⊩□. Let v be an arbitrary world such that wRv. We have v⊩→ and v⊩So, v⊩Generalizing on v, w⊩□→□ 26 Axioms corresponding to properties of accessibility relation R of model M: (□→) corresponds to reflexivity: For any world w, assume w⊩□f R is reflexive, wRw, and w⊩Conversely, if R is not reflexive, it is possible that NOT wRw. So, it is possible to have w⊩□without w⊩ (□→□□) or(D3) corresponds to transitivity: For any world w, assume w⊩□For arbitrary world u such that wRu, we have u⊩. If R is transitive, for any arbitrary world v such that uRv, we have wRv, and v⊩Generalizing, we have u⊩□ Generalizing again, we can conclude w⊩□□Conversely, if R is not transitive, it is possible to have worlds w, u and v such that wRu and wRv but NOT wRu, with w⊩□u⊩but v⊩ Thus, w⊩□□ ( → □□ corresponds to symmetry: Consider a world w, with w⊩Let u be an arbitrary world with wRu. If R is symmetric, uRw and u⊩□Generalizing, w⊩□□Conversely, if R is not symmetric, it is possible to have wRu without having uRw, and so, u⊩□and w⊩□□may be true. (□ → □□ corresponds to symmetry and transitivity: (□ → □□ can be derived from ( → □□and□ → □□ From ( → □□we can derive (□ → □□□ and ( → □□□ From □ → □□we can derive (□□→ □□), and we can conclude (□ → □□ (□(□ → ) → □) or (DL) corresponds to Converse-well-foundedness and transitivity: For any world w, assume w⊩□ Let X be the set of worlds accessible from w in which is false. Since R is converse-well-founded and X is nonempty, there is an element of X, say u, such that for any v, if uRv, v is not in X (u is called R-greatest element). If uRv, by transitivity, wRv. Since v is not in X, v⊩. So, u⊩□and (by the definition of X), u⊩. Thus, u⊩□ → . By the definition of X, wRu. So, w⊩□(□→ →□ To prove the converse, (a) transitivity: 27 from (□(□→)→□we can prove □→□□using necessitation and distribution, as in chapter 1) (b) converse-well-foundedness: Assume the contrary. Then, there exists a non-empty set X with no R-greatest element. Consider a valuation such that for any world u⊩ iff u is not in X. Consider a world w in X (w⊩. For any x in X with wRx, we have x⊩and, xRy for some y in X (and y⊩ So, x⊩□So, x⊩□→ For any z not in X with wRy, we have z⊩and so z⊩□→ So, w⊩□(□ → ). As w is in X, there is an x in X such that wRx (and x⊩So, w⊩□ For any model M with accessibility relation R, M is a model of S4 iff the R is reflective and transitive; M is a model of S5 iff R is reflexive, transitive and symmetric; M is a model of GL iff R is transitive and its converse well-founded. (A converse-well-founded relation cannot be reflexive.) It can be proved that if the logic L (S4/S5/GL) proves A, then A is true in all the worlds of all the models (of L). It can also be proved that if in all the worlds of all (finite) models M, A is true, then the corresponding logic proves it [2]. Some consequences: It can be shown that S4, S5 have only a small set of distinct modalities (sequences of □s, ◊s, and negations, i.e., formulae containing only these connectives)10. This can be derived from their axioms (and can be proved in AproS). But, these can also be explained semantically. GL does not have a small set of distinct modalities. (This can be shown semantically). However, the proofs that are possible - such as the second incompleteness theorem can be explained semantically. Here we give a (mainly) semantic justification for the distinct modalities and some GL proofs. 10 For ease of reading, we do not translate ◊s into □s and negations 28 S5: S5 has a few distinct modalities. We first list these, then show that they are indeed distinct, and then show that there are no other modalities, i.e., all other modalities reduce to one of these. The distinct modalities of S5 are unboxed formulae, formulae beginning with box and formulae beginning with diamond, and their negations (i.e., for the atomic formula P, the distinct modalities are P, □P, ◊P, ~P, ~□P, ~◊P; all other modal formulae involving P (and no other connective other than negation) such as □□P are equivalent to one of these.) To see that the modalities are distinct: P is distinct from □P as P can be true in the current world but false in an accessible world (the converse is false). P is distinct from ◊P since P may be false in the current world but true in some accessible world (the converse is false). □P is distinct from ◊P, as P may be true in only one of multiple accessible worlds (the converse is false). Negations are distinct for similar reasons. To see that these are the only distinct modalities: We show that other modalities reduce to one of these: Repeated □s reduce to one □ (i.e., □□…□P is equivalent to □P). □□…□P implies □P because the accessibility relation is reflexive, and the converse is true since the relation is transitive. Repeated ◊s reduce to one ◊ (i.e., ◊◊…◊P is equivalent to ◊P). ◊P implies ◊◊…◊P because the accessibility relation is reflexive, and the converse is true since the relation is transitive. The modality □◊ reduces to ◊ (i.e., □◊P is equivalent to ◊P). □◊P implies ◊P since the accessibility relation is reflexive, and the converse is true since the relation is transitive and symmetric (i.e., consider a worlds w and v such that wRv and v⊩P, then for all worlds u such that wRu, we will have uRv). The modality ◊□ reduces to □, (i.e., ◊□P is equivalent to □P). □P implies ◊□P since the accessibility relation is reflexive, and the converse is true since the accessibility relation is an equivalence (i.e., consider worlds w and v such that wRv and v⊩P. But, for every world u such that wRu, we also have vRu). Negations can be reasoned similarly. 29 S4: S4 has a few distinct modalities. We first list these, then show that they are indeed distinct, and then show that there are no other modalities, i.e., all other modalities reduce to one of these. Given an atomic formula P, the distinct modalities are: P, □P, ◊P, ◊□P, □◊P, ◊□◊P, □◊□P and their negations. To see that the modalities are distinct: P, □P and ◊P are distinct, as every model of S5 is also a model of S4. ◊□ and □◊ need symmetry to reduce to other modalities, (in a model with asymmetrical accessibility relation, they are different). Symmetry is also needed to reduce ◊□◊ and □◊□. To see that these are the only distinct modalities: We show that the other modalities reduce to one of these. : Repeated □s(◊s) reduce to one □(◊) – the proof given above holds, as it does not need the accessibility relation to be symmetric. □◊□◊P reduces to □◊P: To show that □◊□◊P implies □◊P, note that □◊P implies ◊P due to reflexivity, and this so □◊□◊P reduces to □◊◊P which reduces to □◊P; to see the converse, consider arbitrary worlds u, v, and w such that uRv and vRw such that w⊩P. Then we have w⊩□◊P (if not, there is some world w1 with wRw1 where ◊P is false. But, by transitivity, uRw1. So, □◊P cannot be true at u.) and so, u ⊩□◊□◊P. The next reduction is: ◊□◊□P reduces ◊□P. To see that ◊□◊□P implies ◊□P, consider arbitrary worlds u, v and w such that wRu and uRv and v⊩◊□P. But for every world x such that vRx, we also have wRx. So, w⊩◊□P; To see the converse, consider arbireaty worldsw and u such that wRu and u⊩□P. If u has no accessible worlds other than itself, then if u⊩□P, we also have u⊩ □◊□P, else if there is a world v distinct from u such that uRv, v⊩□◊□P (since by transitivity, v⊩□P and vRv by reflexivity). So w⊩◊□◊□P. 30 GL: GL does not have a small set of distinct modalities because □P is distinct from □□P and so on. This can be seen in a model with two worlds u and v such that uRv. If P is not true at v, □P is false at u, but v⊩□P, since it has no accessible worlds, and so u⊩□□P. (The converse is false, since the accessibility relation is not transitive.) However, there are some reductions possible. Further, some theorems of GL can be explained semantically. Some properties: Unlike S4 and S5, models of GL do not have reflexive accessibility relation. So, it is possible for a world to have no accessible worlds (i.e., no consistent accessible world, as every world is by definition consistent). Such a world can prove □A for any A, including falsum (since □A is provable iff the translation of A is probable in PA, this would imply that PA is inconsistent). The statement of the form ~□A implies that there is at least one accessible (consistent) world, and this corresponds to the consistency of PA. The statement of the form □◊A implies that in every accessible world, ◊A is true. Consider an arbitrary world w with w⊩□◊A, and a world v with wRv. Since v⊩◊A and vRv is false, there has to be a world v1 such that vRv1, with v1⊩A. But wRv1 since the accessibility relation is transitive, and so v1⊩◊A, and as v1Rv1 is false, there has to be a world v2 … Continuing this argument, this branch cannot be finite. So, this model does not have a converse well-founded relation. So, w has no accessible worlds. But in this case, w⊩□Q for any Q. So □◊A implies □Q. Conversely, ◊A implies that there is at least one accessible world, and so □◊A is not provable. 31 2.2 Natural deduction rules The natural deduction rules for the connective □ are: □-E: □ □-I □ These rules, as presented, are not sound. For instance, using the □-I rule presented above, we can prove → □ So, these rules need restrictions. We present restrictions that are directly derived from the semantics. However, we show that these are more suited to the intercalation calculus than to natural deduction. Then, we present Prawitz’ syntactic restrictions (which we motivate using the restrictions we derived semantically). In a modal logical system L, can be derived from assumptions if in every model of L (i.e., a model with the accessibility relation corresponding to L) under every evaluation, is true in every world where is true. Consider a model M of L and world w⊩. If w⊩□, in every world v such that wRv, v⊩. Since we are considering all valuations, has to be entailed by the formulae in v that are forced by w i.e. such that w⊩□. Using this, we can formulate the following rule: “□can be concluded using □I from□i and a derivation of φ from assumptions i” This rule is sound, but not complete for the logics we consider. For instance, we cannot prove (□A→□□ This is because the rule does not consider the properties of the accessibility relation of the logics: If the accessibility relation is transitive, and □A is true in w, then □A and A are true in any accessible world (Let wRv and w⊩□A; for any u, if vRu, then wRu, so, u⊩A. So, v⊩□A). If the accessibility relation is symmetric and transitive, (in addition to what we have above), if □A is true in w, □A is true in any accessible world v. (Let wRv and w⊩□A; for some u, 32 wRu, and u⊩A. Then for any world v such that wRv, we also have vRu. so, v⊩□A). If the accessibility relation is transitive and converse-well-founded, Löb’s rule allows □to be concluded using □I froma derivation of φ from assumption □Using this, we formulate the following rules for the connective □ (□-E corresponds to reflexivity of the accessible relation and is allowed without any restrictions in S4, S5, but not in GL). □-I(S4): □can be concluded using □I from□i and a derivation of φ from assumptions □i11 [□1], [2], □1 … … [□k] φ □k □φ Here, the proof in black can be thought of as corresponding to a world w, and the part of the derivation that is circled in blue can be thought of as a proof in a world v such that wRv. □-I(S5): □can be concluded using □I from□i, ¬□χi and a derivation of φ from assumptions □i, and ¬□χi. □-I(GL): □can be concluded using □I from□i and a derivation of φ from assumptions □i, i and φ(Here we need i as we do not have a separate elimination rule.) 11 i can be derived using □-E. 33 [□1],[1]…[□k],[k],[□φ] □1 … φ □k □φ In the derivation, the part of the derivation that is circled in blue can be thought of as a proof in the “object theory”. Since we have an extra assumption □φ, one proves, really, □φ → φ. Since this implication is “proved in the object theory”, □(□φ → φ) can be concluded. Then, using Löb’s rule, □φ can be deduced. Instead of proving this implication, and using Löb’s rule, the formulation presented here simply allows □φ to be used as an assumption, so the rule reads: □I/E:□can be proved from□i and a derivation of φ from assumptions □i, i and □ These rules are sound and complete, and give normal proofs. However, they but have an unbounded number of premises (e.g., in the figure above, the instance of the inference rule has k+1 premises). This is undesirable for natural deduction, but not a problem in the case of the intercalation calculus that uses a modification of this. So, Prawitz in [7] formulates □-I (for S4 and S5) as follows: 2.2.1 Prawitz’ restrictions on □-I12 Prawitz gives three versions of □-I. Only the last version gives normal proofs. For simplicity, we explain the motivation and the details of the other versions only for S4. 12 Prawitz gives rules for S4 and S5. These do not have restriction on □E. When we give the rules for GL, we give restrictions for □E as well. 34 Version 1: Consider the subderivaton of in the □-I formulated above (the region in blue). All the assumptions are of the form □Using this, a version of the rule can be formulated: if can be proved from □i (for 1 i n) then □ can be concluded from using □I13. This formulation is sound and complete, but does not give normal proofs. E.g.: In proving □(A&B) from □A&□B, to apply the □-I rule on A&B, we have to prove A&B from □A and □B (but not □A&□B). So, we apply →I twice to get □A→(□B →□(A&B)). We prove □A and □B from □A&□B separately and use →E to finish the proof. This proof is not normal. To get around this, Prawitz introduces the concept of essentially modal formulae. If is a formula for which the system can prove → □, then is an essentially modal formula. The S4 formula is an essentially modal formula with respect to S4 if is of the form: iii. □, iv. v. (1 & 2) where 1 and 2 are essentially modal vi. (1 2) where 1 and 2 are essentially modal. Essentially modal formulae with respect to S5 are S5 formulae such that is either: (i) an essentially modal formula with respect to S4 or is of the from , and is essentially modal with respect to S5. Essentially modal formulae with respect to GL are those of S4. Version 2: If can be proved from essentially modal formulae, then □ can be concluded from using □I. Though this version is better than the previous one, it does not always give normal proofs; e.g., To prove □□B from A&□B, a normal proof would use □I with the premise □B to conclude □□B; 13 For S5, we include the negations of such formula as well; for GL, we include □and i. 35 But □B cannot be derived only essentially modal formulae (it can be concluded from A&□B, but this is not essentially modal). So, in this version, we have to prove □□B from □B using the following detour: derive □B→□□B, and derive □B from A&□B, and then use →E to finish the proof. To fix this, version 3 was introduced. Version 3: If Σ is a proof- □ (using □I) along the path such that i. ii. is essentially modal with respect to S4(S5/GL) ψ does not depend on any assumptions χ which φ does not depend on. Note: These restrictions we formulated for S4 and S5, which have a □E rule. In case of GL, we also have the following restriction: On the proof tree described above, □E is allowed on (modal formula) . (□E is allowed only when the conclusion is used in deriving a formula on which □I is to be applied. Since, occurs in what corresponds to an accessible world, this corresponds to the definition of □). 2.3 Intercalation calculus rules These rules can be seen to be similar to the rules for natural deduction that used an unbounded number of assumptions14. 2.3.1 IC rules for S4 □E : ; ? , □δ , ; , δ? Here, we only consider □from . If □is derived using introductions, this can be pushed in inside the current rule, and if it derived using inversion, the current rule can be pushed into it. The details are not given, since we give a completeness proof. 14 36 □I : ; ? □ () ? where () is the set of formulae □ such that □ This is as presented in [10]. 2.3.2 IC rules for S5 □E : ; ? , □δ , ; , δ? □I : ; ? □ () ? where () is the set of formulae γ of the form □ or ¬□ such that γ Note that the rules for S4 and S5 differ only in the definition of (). 2.3.3 IC rules for GL Here, as the natural deduction calculus, we have a □-E/I rule. □E/I : ; ? □ () ? where () is the set of formulae γ of the form iii. □ such that □ , iv. such that □ , v. □ Note that the rules of S4 and GL are similar, though this may not be apparent. There are only two differences: (i) in GL, due to Löb’s theorem, the conclusion of □I can be used as an assumption to derive its premise and (ii) □E rule is combined with □I rule in GL. 37 2.4 Soundness 2.4.1 Soundness proof for of S4 Soundness theorem: If an S4 IC-tree for Г ? evaluates to Y, then in S4, Г ⊩ . Proof: By induction on the height of the IC-trees. The classical rules are dealt with as usual. □I: Let Σ be an IC-tree of height h for from assumptions of Г, and let the last rule applied be □I. So, is of the form □C and the premise of □I is C. Let Σ restricted to C be Σ’. Σ’ is an IC-tree for C from some assumptions Δ; indeed, by the restriction on □I, Δ = . ' C □C Σ′ is of height h-1, and by induction hypothesis, Δ ⊩ C. So, Г ⊩ C. We have to show that Г ⊩ □C. Consider an arbitrary model of Г, M. As Г is a subset of Г, M is a model of Г, and as Г ⊩ C, C is true in any world w of M. Choose any such world u, and for this fixed u, any v with uRv; clearly, both u ⊩ C and v ⊩ C. Thus, u ⊩ □C. As u was an arbitrary world, of M and M an arbitrary model of Г, Г ⊩ □C. □E: Let Σ be an IC-tree of height h for from assumptions of Г, and let the last rule applied be □E. So, is of the form C, and the premise of the □E is □C. Call Σ restricted to C be Σ’. Σ’ is an IC-tree of height h-1 for □C from some assumptions Г. By induction hypothesis, Г ⊩ □C. Let M be an arbitrary model of S4, and u a world in which all the elements of Г are true. Then, u ⊩ □C. As the accessibility relation R is reflexive, u ⊩ C. Hence, Г ⊩ C. 38 2.4.2 Soundness proof for of S5 Soundness theorem: If an S5 IC-tree for Г ? evaluates to Y, then in S5 Г ⊩ . Proof: By induction on the height of IC-trees. The classical rules are dealt with as usual. □I: Let Σ be an IC-tree of height h for from assumptions of Г, and let the last rule applied be □I. So, is of the form □C, and the premise of the □I is C. Let Σ restricted to C be Σ’. Σ’ is an IC-tree for C from some assumptions Δ; indeed, by the restriction on □I, Δ = . ' C □C Σ’ is of height h-1, and by induction hypothesis, Δ ⊩ C. So, Г ⊩ C. We have to show that Г ⊩ □C. Consider an arbitrary model of Г, M. As Г is a subset of Г, M is a model of Г, and as Г ⊩ C, C is true in any world w of M. Choose any such world u, and for this fixed u, any v with uRv; clearly, both u ⊩ C and v ⊩ C. Thus, u ⊩ □C. As u was an arbitrary world of M and M an arbitrary model of Г, Г ⊩ □C. □E: Let Σ be an IC-tree of height h for from assumptions of Г, and let the last rule applied be □E. So, is of the form C, and the premise of □E is □C. Let Σ restricted to C be Σ’. Σ’ is an IC-tree of height h-1 for □C from some assumptions Г. By induction hypothesis, Г⊩□C. Let M be an arbitrary model of S5, and u a world in which all the elements of Г are true. Then, u⊩□C. As the accessibility relation is reflexive, u⊩C. Hence Г⊩C. 39 2.4.2 Soundness proof for of GL Soundness theorem: If an S5 IC-tree for Г ? evaluates to Y, then in S5 Г ⊩ . Proof: By induction on the height of the IC-trees. The classical rules are dealt with as usual. (i)Let Σ be an IC-tree of height h for from assumptions of Г, and let the last rule applied be □I. So, is of the form □C, and the premise of the □I is C. Let Σ restricted to C be Σ’. Σ’ is an IC-tree for C from some assumptions Δ; indeed, by the restriction on □I, Δ = . ' C □C Σ’ is of height h-1, and by induction hypothesis, Δ ⊩ C. So, Г ⊩C. We have to show that Г ⊩ □C. Consider an arbitrary model of M. Consider an arbitrary world u which is a model of Г. Let v be a world such that uRv. Since Г is true in u, for any formula of the form □χ that is contained in Г, χ is true in v. Since M is a GL model it is transitive. So, □χ is also true in v. Г ⊩C. By definition of Г, v ∪ {□C} = Г. So, v ∪ {□C} ⊩C or v ⊩ (□C→C) Since v is an arbitrary world accessed by u, generalizing on v, u⊩□ (□C→C). By Löb’s rule, u⊩C. Since u is an arbitrary world which makes all the elements of Г true, Г⊩C. As elimination rule is combined with introduction rule, the same reason holds for that as well. 40 2.5 Completeness The completeness proofs for all three systems proceed along the same lines. The proof for S5 is presented first as it is the easiest. The proofs of the other systems differ slightly from that of S5, and only these differences are presented. 2.5.1 Completeness of S5 rules Completeness Theorem: Either the S5 IC-tree for ? G contains a normal S5 derivation for ? G or it allows the definition of a counterexample to the inference from to G. This is proved directly from the proof extraction theorem and counterexample extraction theorem proved below. Proof Extraction Theorem: For any and G, if the S5 IC-tree for β ? G evaluates to Y, then a normal nd-proof of G from the assumptions in can be found. Proof: We prove this by first showing that an IC-derivation D can be extracted from the IC-tree T for β ? G evaluates to Y, and then showing that from any IC-derivation D, a normal ndproof of G from can be extracted. To show that an IC-derivation D can be extracted from the IC-tree T for β ? G evaluates to Y: An IC-derivation of G from α is a subtree D of the IC-tree T for 〈α β ? G〉 such that 〈α β? G〉 is the root of D, all the branches of D are Y-closed branches of T, and every question node (node corresponding to a question) in D that is not the root is followed by exactly one rule. Note that since T is finite, and every leaf node ends with a Y or an N, this assignment can be propagated down to all the nodes in the obvious way. Define D as f(height(T)) where f is defined as: f(0) = 〈α β ? G〉; f(2n) = g1(f(2n)); f(2n+1) = g2(2n) 41 where g1(x) is defined as the leftmost rule application that extends the derivation, i.e., has all of its premises evaluating to Y if one exists, and x otherwise, and g2(x) returns the appropriate questions for the rule chosen by g1(x) if g1(x) is a rule, and x otherwise. It is obvious that D is an IC-derivation. To show that from any IC-derivation D, a normal nd-proof of G from can be extracted: The proof proceeds by induction on height(D). The base case is where height(D)=1. In this case, D consists of just one question β ? G where G is in β. The nd-proof for this is just the node G. If height(D) = h (> 1), the proof proceeds in cases depending on the last rule application used, with the induction hypothesis that states that given any IC-derivation T such that height(T) < h, a normal nd-proof can be extracted from it. The propositional rules are dealt as in [10]. For the modal rules, □I: If D is an IC-derivation for β ? G such that its last rule application is □I, then the immediate subderivation of D, say D’ is an IC-derivation for β) ? Hwhere G = □H. By induction hypothesis, we can extract a normal nd-proof P from D’. Construct P as follows: The root of P is G and the immediate subproof if P is P’ and the last inference rule used is □I. P is a normal nd-proof that is associated with D. Since D’ used β) as it premises, the restriction for □I in nd proofs is satisfied (the formulae in are β)the essentially modal formulae that occur from H to the assumptions it uses from β. □E: If D is an IC-derivation for β ? G such that its last rule application is □E, then the immediate subderivation of D, say D’ is an IC-derivation for β ? Hwhere H = □G. By induction hypothesis, we can extract a normal nd-proof P from D’. Construct P as follows: If P’ contains occurrences of H as open assumptions, then replace them by the inference rule □E with its premise being H and conclusion being G. P is a normal nd proof that is associated with D. 42 nd-proofs extracted from IC-derivations are normal, because the elimination rules are applied from above whereas the introduction rules are applied only from below. Counterexample Extraction Theorem: For any and G, if the S5 IC-tree for ? G evaluates to N, then using this tree, it is possible to construct an S5 model in which all the elements of are true but G is false. Proof: We want to construct a world w in an S5 model M where is true but G is not. We construct an S5 model, i.e., a nonempty set W (with w), a binary equivalence relation R and the relation ⊩. The following describes the construction of each of these. Structure of the set W: Since there are only finitely many formulae involved, and inaccessible worlds do not influence each other, a counter example can be found in a finite model. Accessibility relation R: every world in W is accessible to every other world. Construction of the set W: One of the worlds, w, makes true. The other worlds are constructed using the fact that □I rule uses the idea of accessible world – if formula F is allowed as an assumption to prove a formula A to which □I is applied, F is true in all accessible worlds. Thus, applying □I rules to all formulae to which this can be applied will generate all possible accessible worlds. The truth of a formula depends only on the truth of its subformulae (in all accessible worlds), and we need to construct only a finite number of worlds with all possible valuations of subformulae. This is done as follows: A subtree of the IC-tree is constructed as follows: Select a single branch P0, all of whose nodes evaluate to N (using rules). Apply □I to all nodes of P0 to which this rule is applicable.15 This gives the root node of a new branch16. 15 Except for the root – reason for this is in footnote 7. 43 Repeat the process till no new branches can be obtained. As the whole IC-tree is finite, this procedure halts and returns a subtree. The top nodes of a branch Pi of this subtree is a question of the from (i, ? Gi). Let Ai be the elements in i and Gi- (negation of Gi). For each branch Pi, Ai is added to the set W. A0 corresponds to w. Relation R: For every i, j Ai, is related to Aj; ⊩ the standard valuation relation17. (Detailed) Proof: Define + and - as follows: + = if = and + = otherwise. - = if = and - = otherwise. Enumerate F(, G) by Hi, 0 i n. F(x) is the set of all unnegated proper subformulae of formulae in x and the unnegated part of all negations which are subformulae of formulae in x. Assume the IC-tree for ( ? G) evaluates to N. Construction of W: Construction of the subtree: Construction of P0: The sequence of nodes P0*(0),… is defined as follows: Let 0 = , 0 = 0, G0 = G, H0 = G. m+1 is defined according to the following cases. Case1: (j)[ (m j n) & (Hj is not of the form □) & ((Hj m) & (Hj m))] . 16 For S5, we only consider the last branch got by □I – this is because the accessible world has formulae that are negated modal as well. 17 To ensure that every branch gives a world that respects the definition of □, we choose P0 as follows: There are several formula pairs on which can be applied; we choose the branch that considers the boxed formulae last (thus all relevant formula are already incorporated into the branch’s premises) 44 Then m+1 is the least such j. Case2: The previous case does not apply, (j)[ (m j n) & (Hj is of the form □) & ((Hj m) & (Hj m)) & (m ? Hj evaluates to N)]. Then m+1 is the least such j. Case3: The previous cases do not apply, (j)[ (m j n) & (Hj is of the form □) & ((Hj m) & (Hj m)) & (m ? Hj evaluates to Y)]. Then m+1 is the least such j. Case4: the previous cases do not apply. Then let m+1 = 0. Then, let Gm = Hm Gm = Hm if m ? Hm evaluates to N, otherwise m+1 = m, G-m P0*(2m) = m ? Gm P0*(2m+1) = i, Hm+1 if Gm is a negation c, Hm+1 otherwise Let be the smallest m with m+1 = 0. 45 Define P0 to be P0* restricted to {m | m 2}. Construction of remaining branches: Now, consider the nodes of the form ’ ? □ in P0, (excluding the root). These nodes appear in P0 only because of case 3, hence only after all the formulae of F in (, G-) not of the form □ have been used. To each of these nodes, the rule □I is applicable, leading to a node of the form ? (which evaluates to N). Note that contains all the formulae of the form □ψ in , which are obtained using case 3. So, contains all formulae of the form □ψ obtained using case 2, and all the formulae of the form □ψ obtained using case 3. Start the construction of the branch P0, choosing at each stage the following node according to the cases for P0. Then repeat the process for each of the branches. For each Pi18, let ’i = i?Gi be the top node of Pi. Define Ai = { | i, Gi- }. Let W be the set of all Ai Lemma: For 0 i, j r, the following claims hold: 1. if Ai, then - Ai 2. if is a subformula of an element of an element in Ai, then either + Ai or - Ai 3. if Ai, then Ai, 4. if (1 & 2) Ai , then 1+ Ai and 2+ Ai 5. if (1 & 2) Ai, then 1- Ai or 2- Ai 6. if (1 2) Ai, then 1+ Ai or 2+ Ai 7. if (1 2) Ai, then 1- Ai and 2- Ai 8. if (1→ 2) Ai, then 1- Ai or 2+ Ai Since we want formulae of the form □and □ in the accessible world, we only consider the last □I applied in a branch. This means all the s4 worlds will have the same modal formulae (and their negation), which is a property of the system. In S4 and GL, every application of □I results in a new branch. 18 46 9. if (1→ 2) Ai, then 1+ Ai and 2- Ai 10. if □ Ai, then for every j, j contains + 11. if □ Ai, then there is an r such that j r and r which contains -. Proof: proof for (1) – (9) is the same as classical logic. (10): if □ Ai, then □ must appear on the left side of the question mark below any node is put in W. This is because of the ordering of the cases. The formulae of the form □ are never taken out of the left side of the question mark. On application of □I, for all the formulae of the form □ and the corresponding s are transferred into a new branch. Thus, all the branches thus obtained will contain □ and . Thus all worlds Aj such that i<j, Aj contains +. Further, consider world Ak such that k∈i. If Ak contains -, then Ak does not contain □ (or using □E, we can get a contradiction). Hence, Ak contains +. (11): if □ Ai,, then □ has been dealt with in case 3. Thus, a new node j = ’ ? has been placed in W, and since the rule applied to j is either i or c, - appears on the left side of the question mark in Pj, so - Aj. Hence, the ⊩ relation holds and so this provides a counter example. Hence, either the S5 IC-tree for ? G contains a S5 derivation for ? G or it allows the definition of a counterexample to the inference from to G. 2.4.1 Completeness of S4 rules Completeness Theorem: Either the S4 IC-tree for ? G contains a S4 derivation for ? G or it allows the definition of a counterexample to the inference from to G. 47 This is proved directly from the proof extraction theorem and counterexample extraction theorem proved below. Proof Extraction Theorem: For any and G, if the S4 IC-tree for ? G evaluates to Y, then a normal nd-proof of G from the assumptions in can be found. Proof: The proof that a normal nd proof can be extraction from any IC-tree that evaluates to Y proceeds exactly the same as the proof for S5. nd-proofs extracted from IC-derivations are normal, because the elimination rules are applied from above where as the introduction rules are applied only from below. Counterexample Extraction Theorem: For any and G, if the S4 IC-tree for ? G evaluates to N, then using this tree, it is possible to construct an S4 model in which all the elements of are true but G is false. Proof: Similar to the proof above, with the following modifications: Structure of the set W: Since there are only finitely many formulae involved, and inaccessible worlds do not influence each other, a counter example can be found in a finite model. Accessibility relation R: The accessibility relation for S4 is transitive and reflexive. So, any model of S4 can be folded out as a tree (this may involve two worlds being identical in the tree). The worlds are constructed out of the branches of the proof tree. The accessibility relation R: if there is a (possibly empty) path from branch x to branch y, then world corresponding to y is accessible from the world corresponding to x. This is implemented as explained below: Branch Pi,j: jth branch constructed (running count for all the branches). j is the branch number and i is the list of the branch numbers of the branch’s ancestors (branches containing its root node.) Note that i includes j for S4. The proof proceeds very similar to that of S5 except for the following differences: 48 (a) As mentioned, branches are considered at every application of □I, and a single branch has two indices. Thus, construction of the main branch and the branches constructed from the main branch differ from that of S5 branches – though mainly in their names. (b) Relation R: Ai,j is related to Ak,l iff jk (where Ai,j is obtained from Pi,j in the same manner that Ai was obtained from Pi in the proof for S4). In the proof of the lemma, (10) sub proof is (10) if □ Ai,j, then for each r such that j r, r,k contains + Proof: if □ Ai,j, then □ must appear on the left side of the question mark below any node is put in W. This is because of the ordering of the cases. The formulae of the form □ are never taken out of the left side of the question mark. On application of □I, for all the formulae of the form □ and the corresponding s are transferred into a new branch. Thus, all the branches thus obtained will contain □ and . Note that the modification of the accessibility relation allows for a counter example to be constructed for questions like ◊P ? □◊PThe counterexample constructed in this case will contain a world where P, ◊P are true and an accessible world where P is not true. Such a model is possible in this accessibility relation is an equivalence relation. 2.5.3 Completeness of GL rules Completeness Theorem: Either the GL IC-tree for ( ? G) contains a normal GL derivation for ? G or it allows the definition of a counterexample to the inference from to G. This is proved directly from the proof extraction theorem and counterexample extraction theorem proved below. 49 Proof Extraction Theorem: For any and G, if the GL IC-tree for ? G evaluates to Y, then a normal nd-proof of G from the assumptions in can be found. Proof: The extraction of a normal nd proof from an IC-tree that evaluates to Y proceeds similar to the proof of S5, but with some differences. An IC-derivation can be extracted from an IC-tree that evaluates to Y using the procedure given above. To prove that it is possible to extract a normal nd proof from thie IC-derivation D: The proof proceeds by induction on height(D). The base case is where height(D)=1. In this case, D consists of just one question β ? G where G is in β. The nd-proof for this is just the node G. If height(D) = h (> 1), the proof proceeds in cases depending on the last rule application used, with the induction hypothesis that states that given any IC-derivation T such that height(T) < h, a normal nd-proof can be extracted from it. The propositional rules are dealt as in [10]. For the modal rules, □I: If D is an IC-derivation for β ? G such that its last rule application is □I, then the immediate subderivation of D, say D’ is an IC-derivation for β) ? Hwhere G = □H. β)contains the formulae of the form □ and G where □is in β). Let E denote the set of formulae as above. By induction hypothesis, we can extract a normal nd-proof P from D’. Construct P as follows: The root of P is G and the immediate subproof if P is P’ and the last inference rule used is □I. Further, if P has occurrences of formulae □in E that are open assumptions, replace them with the inference rule □□P is a normal nd-proof that is associated with D. Since D’ used β) as it premises, the restriction for □I in nd proofs is satisfied (the formulae in are β)the essentially modal formulae that occur from H to the assumptions it uses from β. 50 nd-proofs extracted from IC-derivations are normal, because the elimination rules are applied from above where as the introduction rules are applied only from below. Counterexample Extraction Theorem: For any and G, if the GL IC-tree for ? G evaluates to N, then using this tree, it is possible to construct an S4 model in which all the elements of are true but G is false. Proof: The proof is similar to that for S4. Structure of the set W: Since there are only finitely many formulae involved, and inaccessible worlds do not influence each other, a counter example can be found in a finite model. Accessibility relation R: The accessibility relation for GL is transitive and converse-wellfounded. So, any model of GL that with an acyclic transitive relation can be a model of GL. In particular, any model can be folded out as a tree (this may involve two worlds being identical in the tree). GL’s models have accessibility relations that are converse well-founded and transitive. Any model of GL can be folded out as a tree (this may involve two worlds being identical in the tree) with the criteria that no node can access itself. Since we are dealing with a finite tree, the model is converse well-founded. This is implemented as follows: if there is a (nonempty empty) path from branch x to branch y, then world corresponding to y is accessible from the world corresponding to x. Branch Pi,j: jth branch constructed (running count for all the branches). j is the branch number and i is the list of the branch numbers of the branch’s ancestors. Note that i does not include j. The proof proceeds very similar to that of S4 except for the following: In S4, the branches were numbered such that each branch included itself in listing its ancestors, while in GL, a branch is not considered its own ancestor. This is because the accessibility model of S4 is reflexive, while that of GL is not. 51 This allows to get counterexample for questions like □A ? AA counterexample for this has a world where A is not true. Since accessibility relation is converse well-founded, there can be no closed path of accessible worlds, and in presence of transitivity, this means that no world can access itself. If this were not true, then the world u can access itself, and so □A is true vacuously. Since the world in inaccessible to itself, the boxed formulae that are true in it do not influence any other formulae in it. Another instance of a counterexample is for □(□A→ A)A counterexample for this has a world w with an accessible world u such that A is false in u. Since u does not have any accessible worlds (in particular, it does not access itself), □A is true in u, but A is not, and so □A→ A is false in u and so □(□A→ A) is false in w. Note that the sentence □(□A→ A) is true in S4 and S5. Further, it is possible in S4 to have a counterexample for the question □(□A→ A) ? □A This will have a world w where A is false. This would mean that □A is false, and so (□A→ A) is true. As it is the only accessible world. □ (□A→ A) is also true, but as shown □A is false. This will not be possible in the case of GL, since no world can access itself. 52 CHAPTER 3 Implementation Proof search procedures for S4, S5 and GL were implemented in the automated theorem prover AProS [1]. This chapter discusses the details of this implementation. The first section describes AProS - in particular the existing proof search procedure for classical sentential logic, the second section describes the additional functionalities that need to be added to this proof search procedure in order to do proof search in modal logic, and the third section explains how these functionalities are implemented and incorporated into the existing system. 3.1 AProS: AProS (Automated Proof Search) is a theorem prover that uses the intercalation method to search for normal natural deduction proofs in sentential and predicate logic in classical, intuitonistic and minimal version. Since S4, S5, and GL are based on classical sentential logic, a discussion of the sentential component of AProS is sufficient to explain the implementation of proof search for these systems. AProS takes as input an assertion (a set of premises and a conclusion) and tries to find a proof for this assertion. Since AProS uses the intercalation calculus, it finds only normal proofs, and its proofs have the subformula property. This helps in constraining the space for proof search. AProS has an internal representation of the IC tree as a tree of occurrences, where each occurrence is of the form 〈F, S〉 where F is a formula, and S is its scope (that is, the premises and assumptions accessible to this occurrence). A natural deduction proof can be directly derived from this tree 53 through a process of enumeration in which the proof is constructing by traversing the tree depth first starting with the goal. The space of proofs (trees) is searched (in depth first manner) guided by strategic steps as explained below. A single tree is maintained and modified as the search proceeds. The most distinctive feature of the proof search procedure implemented in AProS is that the proof construction can be separated strategically into three distinct modules: extraction or goaldirected forward use of elimination rules, inversion or backward use of introduction rules, and finally the use of indirect argumentation. AProS (roughly) does the following: 1) Given an assertion, AProS initializes a partial-proof-tree with it. A partial-proof-tree is a tree whose nodes are occurrences as described above (i.e., containing a formula and its scope). Throughout the proof search, AProS maintains and modifies a single tree. A question node of the form 〈α; β? G〉 in an IC tree is represented in the partial proof tree as a node (an occurrence) of the form 〈F, S〉 where F is the formula of the occurrence (i.e., F is G) and S is the scope of this formula, i.e., the set of assumptions available to it (i.e., S is α); the set of formulae that are extracted from α (i.e., β) is not stored explicitly anywhere, but can be calculated (as explained in (2)). 2) AProS checks if the current goal can be obtained via a sequence of elimination rules starting with the positive occurrences of the premises or assumptions. (Such a sequence is called an extraction sequence.) If so, AProS adds the last rule in the extraction sequence to the tree. The major premise of the rule will be the next rule in the extraction sequence or an assumption, but the rule may have minor premises which need to be proved, and such a premise is added to a list called the list of open-questions. This process is repeated for each rule in the extraction sequence. Then, AProS marks an open question as the current goal and searches for a proof using steps (2), (3) or (4). This is repeated till no open questions are left. If this succeeds, the goal is said to be extractable from the premises. If no proof is found for an open question, AProS 54 backtracks, (as in step 5) and searches for other extraction sequences that end with the current goal, and repeats this procedure using them19. 3) If the current goal is not extractable, AProS checks if the current goal can be obtained as a conclusion of an introduction rule. If so, the tree is updated by adding the introduction rule application above the current goal. This adds the premises of the rule to the tree. If a premise is not in the set of premises and assumptions available to it is added to the list of open questions, and this list is dealt with as above. 4) If steps (2) and (3) fail, (i.e., if the current goal G is not extractable, and is not a conclusion of the introduction rule, or if the open questions generated above the current goal in steps (2) and (3) cannot be proved, and AProS backtracks to G) the current goal is negated and added to the available assumptions20. Then the algorithm tries to find a contradiction to use in an indirect argument. This is done as follows: AProS generates a list of negated formulae that are positively contained in the premises or the assumptions on which G depends, then generates a list of contradictory pairs by using these formulae and their immediate subformula. AProS then adds to the partial-proof-tree, a negation rule (introduction or elimination depending on the current goal) and a falsum rule above it. Each pair from the list of contradictory pairs is added to as premises to the falsum rule one by one till either the list is exhausted or a pair thus added is proved successfully (i.e., both the formulae are added to the list of open questions, and both the open questions are eventually added to the tree as goals, and proved). 5) If the current goal is such that no rules can be applied, or steps (2), (3) and (4) have been applied to this goal, but have failed, or is a repeated question, (i.e., during the course of the search, AProS has tried to prove this goal using a superset of the assumptions currently available to it, and failed) then AProS retracts its goals in a depth first order, marking that the current goal AProS treats as a special case the extraction sequences of length 1 - i.e., when the goal is in the scope. Here, it applies the “premise rule” which simply closes the current branch with a premise. 20 If G is of the form F, its negated version is simply F. 19 55 cannot be proved from the assumptions available to it. As mentioned in the steps, if it is possible to proceed in more than one way in any step, then AProS does so in depth first manner. If every step has succeeded, (there are no open questions left to be proved) then the tree is a fully justified tree, from which a proof can be extracted. If however, the root of the tree fails (as in step 5) then the proof search fails. Note that though a question node differs from an occurrence in that the set of formulae that are extracted from α (i.e., β) is not stored explicitly anywhere, but is calculated in a lazy fashion using extraction, the semantics of this match the eager calculation of β since extraction is done before any other strategy is used, and at each step AProS proceeds in a depth-first manner so that after each rule is added (by any step), extraction is tried first. 3.2 Proof search in modal logic This section describes the additional functionalities that need to be added to AProS in order to implement the proof search for modal logic. The implementational details are given in the next section. The rules are restated here. IC rules for S4/S5 □E : ; ? , □δ , ; , δ? □I : ; ? □ () ? where () is in S4 the set of formulae □ such that □ and in S5 the set of formulae γ of the form □ or ¬□ such that γ IC rules for GL □E/I : ; ? □ () ? where () is the set of formulae γ of the form 56 vi. □ such that □ , vii. such that □ , viii. □ The most direct translation into AProS that will add the required functionalities is given here. As before, AProS initializes a partial proof tree of occurrences with the goal. As explained in the previous chapters, modal logic introduces a new unary connective - □ (◊ is defined as in terms of □). So, the main changes include adding inference rules for the □, and modifying the proof search to incorporate this. Additions to the available rules: 1. In S4/S5, □E is added as an elimination rule – it is used as any other elimination rule. 2. GL has no □E rule. 3. □I is added as an introduction rule, but to satisfy its restrictions the following changes are needed: when □I rule is applied to an occurrence 〈□F, S〉, the premise of that rule that is added as an open question to the partial proof tree is the occurrence 〈□F, S’〉 where S' corresponds to (), i.e., for S4/S5, S' is all boxed formulae in S that are extractable from S, and for GL, it is all the boxed formulae in S and extractable from S, their immediate subformula, and the conclusion of □I. Modifications to the proof search: 1. Construction of S’ would involve listing all the boxed formulae that are extractable, i.e., an eager evaluation of the extractable fromulae. AProS does not keep an explicit list of all the formulae that are extracted, or that are extractable, but computes this in a lazy manner. So, the proof search has to be modified to include eager evaluation. This would involve either (a) using multiple partial-proof-trees or (b) modifying the search algorithm to deviate from depth first search (if an open question fails, AProS backtracks and fails the rule that generated it, but if AProS checks a formula and finds it unextractable, instead of backtracking it should move on to the next potential formula in order to list all the extractable formulae.) 57 2. Although conceptually simple, changing the scope of an occurrence so that it does not contain the scope of the occurrence immediately below it would cause some implementational changes – especially in printing the proofs. 3.3 Implementational details The modifications to the proof search procedure listed above cannot be incorporated easily into the existing implementation. So, we modify the algorithm presented above to use lazy search so that while it preserves the semantics of the functionalities needed, it needs fewer modifications to the existing proof search procedure. We compute the extractable boxed formulae in a lazy manner that can be incorporated seamlessly into the lazy computation of extractable formulae by the existing proof search. In this approach, instead of listing 〈() ? P〉 and trying to prove P from it, we allow (existing) proof search algorithm to try to prove P from α. But, when the proof search algorithm tries to use an occurrence of a formula from α, we check if it satisfies the restrictions, i.e., that the occurrence is indeed used to extract a formulae that will be in () called the allowed-modal formula (boxed formulae in case of S4/S5 and boxed formulae, their unboxed versions and the conclusion of the □I rule in the case of GL). In particular, all the extraction sequences have to satisfy this restriction, i.e., contain a boxed formula. If not, we disallow the use of that occurrence, i.e, if AProS attempts to extend the subtree by adding a node that would result in violating the restrictions, this attempt fails, and the algorithm backtracks in the usual depth first fashion. So, the only addition we have to the existing algorithm is to reject a proof of P that does not meet the restrictions. (For GL, the allowed-formulae contain □P, but it may not be extractable from α, so we explicitly add □P to the scope of P before proceeding to attempt to prove P.) The following are the functions that were implemented to carry out the proof search. The list below gives the functionality of each function instead of the algorithm. 58 Allowed-modal formula: This function checks whether a given formula is an allowed-modal formula in the modal logic that is currently being used. This is used to check if an occurrence of the formula present in the available assumptions can be used to extend the tree. For GL, an additional function checks if the given formula is the conclusion of a □I rule (if so, the formula can be justified by Löb’s rule). Depends upon: This function returns the set of occurrences that a given occurrence depends upon. This set is different from all the assumptions available to the given occurrence, as some assumptions may not be used in the tree for proving the occurrence. This is also different from the set of top nodes of the subtree defined by the occurrence, as some of the top nodes may be assumptions that are introduced and discharged above the given occurrence. Can use assumption: This function checks whether using an occurrence of an available assumption to extend the subtree violates the restrictions for the □-rules. This is done as follows: For every application of □I (below the goal), the following are checked: 1. If the assumption is available to the conclusion of a □I-rule, it should be used to obtain an allowed-modal formula through extraction. 2. If the assumption is not available to a □-rule21, then no allowed-modal formula that was needed by another assumption should depend on this. GL □E: In GL, □E is allowed only on allowed-modal formulae that are extractable from the available assumptions for some application of □I, such that it depends on only these assumptions. Note: 21 this is possible since the assumption may be introduced after the □I rule was applied, and its addition may have nothing to do with the □I. 59 The rules allowed in extracting does not include □E. This function checks if the application of a □E satisfies these restrictions. 3.4 Example: Assertion in S4 Set of premises: {□(A B), A, □A} Goal: □B. The following shows a proof in IC {□(A B), □A} ; {(A B),A} ? A (□E) {□(A B), □A} ; {(A B, B)} ? B {□(A), □A} ; {(A B)} ? A (E) □(A B), □A} ; {(A B)} ? B (□E) {□(A B), □A} ; {} ? B (□I) (A is boxed and so excluded) {□(A B), A, □A} ; {} ? □B The following shows the tree as it evolves in AProS: Tree 1: In this step, a tree with just the root node is created and initialized with the given question as occurrence (〈□F, S〉 where F is a formula and S, a set of formulae, is its scope). The current goal is marked with an “*”. 〈□B, {□(A B), A, □A}〉* Tree 2: Since the goal is not extractable, but can be obtained as a conclusion of an introduction rule, the proof search attempts this. Since the proof search uses lazy evaluation of the extractable formulae, all the premises are carried over. 〈B, {□(A B), A, □A} 〉* □I 〈□B, {□(A B), A, □A}〉 60 Tree 3: The goal is extractable with the extraction sequence: □(A B) A B B In the path from the assumption used to the premise of □I, (B), there is a boxed formula □(A B). Hence the restriction is satisfied and the extraction sequence is allowed. The proof search procedure then proceeds to add the last element of the sequence to the tree. 〈A, {□(A B), A, □A}〉* 〈 (A B), {□(A B), A, □A }〉 E 〈B, {□(A B), A, □A} 〉 □I 〈□B, {□(A B), A, □A} 〉 Tree 4: The current goal is its scope. However, this premise does not satisfy the restrictions of □I rule. Hence, this fails. PREMISE -FAIL 〈A, {□(A B), A, □A}〉 〈 (A B), {□(A B), A, □A} 〉 E 〈B, {□(A B), A, □A} 〉 □I 〈□B, {□(A B), A, □A} 〉 61 Tree5: Since the proof search procedure backtracks in a depth first manner, it tries to prove the same goal using other means. The goal is extractable by the following extraction sequence AA This extraction sequence satisfies the restrictions, and so is tried. 〈□A, {□(A B), A, □A}〉* □E 〈A, {□(A B), A, □A}〉 〈 (A B), {□(A B), A, □A}〉 □E 〈B, {□(A B), A, □A}〉 □I 〈□B, {□(A B), A, □A} 〉 Tree6: The current goal is in the scope. Further, it is adding the premise does not violate the restrictions of □I. So, this branch succeeds. PREMISE - SUCCEED 〈□A, {□(A B), A, □A}〉 □E 〈A, {□(A B), A, □A}〉 〈 (A B), {□(A B), A, □A }〉* E 〈B, {□(A B), A, □A}〉 □I 〈□B, {□(A B), A, □A} 〉 62 Tree 7: This question was part of the original extraction sequence □(A B) A B B So, the proof search procedure simply continues by adding the next element. PREMISE - SUCCEED 〈□A, {□(A B), A, □A}〉 〈 (A B), {□(A B), A, □A }〉* □E □E 〈A, {□(A B), A, □A}〉 〈 (A B), {□(A B), A, □A }〉 E 〈B, {□(A B), A, □A}〉 □I 〈□B, {□(A B), A, □A}〉 Tree 8: The current goal is in the scope, and can be added without violating any restrictions. PREMISE - SUCCEED PREMISE - SUCCEED 〈 (A B), {□(A B), A, □A}〉 〈□A, {□(A B), A, □A}〉 □E □E 〈 (A B), {□(A B), A, □A}〉 〈A, {□(A B), A, □A}〉 E 〈B, {□(A B), A, □A}〉 □I 〈□B, {□(A B), A, □A}〉 This completes the proof search which was successful. 63 This example illustrates how the lazy evaluation of the extractable boxed formulae works. Above a □I rule, an extraction sequence is allowed only of it contains a boxed formula, and a premise is allowed to be added as an initial rule only if a boxed formula is extracted from it (above the □I rule). The example also shows how the proof search proceeds in a depth first manner – i.e., when one of the “proofs” of a question violates the restriction, the proof search backtracks and tries to prove it again. This can be incorporated easily into the existing proof search since back tracking is done as a matter of course (for instance, when one of the open questions corresponding to a minor premise of an elimination rule cannot be proved). 64 Appendix This section presents a list of theorems that AProS proves (or in unable to prove) in the modal logics S4, S5 and GL. We include the proofs (IC-derivations) of a selected few22. The list of modal logical statements was generated as follows: 1. Axioms of a logic are provable in it. Axioms of one logical system L1 that are not provable in another logic L2 are listed under unprovable statements of L2. 2. Properties that the models of a logic should satisfy can be proved in it. Properties that the models need not satisfy are unprovable in it. 3. Distinct modalities: A sequence of boxes and negations is called a modality. For a given logic, there is a set of modalities DS such that for any other modality, there is some modality in DS that it is equivalent, and no two modalities in DS are equivalent to each other. DS is called the set of distinct modalities. A logic cannot prove that any two elements in the set are equivalent. When DS is finite, all such statements are shown to be unprovable. When DS is infinite, a few of these are shown. Further, other modalities that are equivalent to ones in DS follow some pattern, called the reduction rules. We prove the reduction rules and give examples of application of the rules. 4. Finally, we list a set of miscellaneous examples obtained by modifying one of the examples listed above to include various connectives. 1. Examples in S4: Theorems: The examples are organized as follows: I. II. III. IV. Axioms of S4 Properties of S4 models Reduction rules. Miscellaneous examples with other connectives I. Axioms of S4. Necessitation: For any tautology A, S4 ⊦ □A 1. S4 ⊦ {} ? □(A→A) Y {A} ? (A) I {} ? (AA) □I {} ? □(AA) 22 Some branches in some derivations leave out unwanted or repeated assumptions for the sake of space, 65 2. S4 ⊦ {} ? □(A∨ ¬A) 3. S4 ⊦ {} ? □(P→(Q→P)) Distribution 4. S4 ⊦{□(R→S), □R} ? □S Y Y {; RS,R,S} ? S {; RS,R} ? R EE {; RS,R} ? S □E {□R; RS } ? S □E {□(RS),□R} ? S □I □(AA) {□(RS),□R} ? □S □(AA) 5. S4 ⊦ {□A} ? A Y {;A} ? (A) □E {□A} ? A 6. S4 ⊦ {□P} ? □□P Y {□A}?□A □I {□A}?□□A 66 Axioms of S5 corresponding to symmetry, Löb axiom are unprovable, and AProS shows this. II. Properties of S4 models. From definition of □ 7. S4 ⊦{□(P & ¬P)} ? □Q 8. S4 ⊦{□P, ¬□¬¬P} ? Q 9. S4 ⊦{□¬P, ¬□¬¬¬P} ? Q The proofs are direct and not shown here. Reflexivity, transitivity are directly axioms. Symmetry and well-foundedness are not required properties of the accessibility relation of S4 models, and so are unprovable. AProS shows that they are unprovable. III. Distinct Modalities The set of distinct modalities for S4 are: DS = {*P, P | * is one of □, ◊, ◊□, □◊, ◊□◊, □◊□, ¬□, ¬◊, ¬◊□, ¬□◊, ¬◊□◊, ¬□◊□, ¬} All the other modalities reduce to one of these using the following reduction rules. Reduction rules (proved in the next pages) 10. S4 ⊦{} ? □P ↔ □□P, (shown above) 11. S4 ⊦{} ? ¬□¬P ↔ ¬□¬¬□¬P i.e. {} ? ◊P ↔ ◊◊P 12. S4 ⊦{} ? □¬□¬□¬□¬P ↔ □¬□¬P i.e.{} ? □◊□◊P↔ □◊P 13. S4 ⊦{} ? ¬□¬□¬□¬□P ↔ ¬□¬□P i.e.{} ? ◊□◊□P ↔ ◊□P Since the rules listed here are bi-implications, the negated versions of these rules are already proves. However, AProS was made to prove them since the logics differ in the way their boxrules handle negations and these verified this interaction. The proofs of these are not listed. 14. S4 ⊦{} ? ¬□P ↔ ¬□□P, 15. S4 ⊦{} ? ¬¬□¬P ↔ ¬¬□¬¬□¬P, i.e.{} ? ¬◊P ↔ ¬◊◊P 16. S4 ⊦{} ? ¬□¬□¬□¬□¬P ↔ ¬□¬□¬P i.e.{} ? ¬□◊□◊P ↔ ¬□◊P 17. S4 ⊦{} ? ¬¬□¬□¬□¬□P ↔ ¬¬□¬□P i.e.{} ? ¬◊□◊□P ↔ ¬◊□P Modal logics are usually presented using connectives □ and ◊. Because of this, the formulae ¬□P and ◊¬P are syntactically different, and reduction rules list them. Since we only use □, ◊¬P translates to ¬□¬¬P. Proving these two are equal is now simply a propositional proof. However, 67 AProS was made to prove these since interaction between negation rules and box rules have to be verified. We do not give the proofs of these here. 18. S4 ⊦{} ? (□¬P ↔ ¬¬□¬P), i.e. {} ? (□¬P ↔ ¬◊P) 19. S4 ⊦{} ? (¬□¬¬P ↔ ¬□P), i.e.{} ? (◊¬P ↔ ¬□P) 20. S4 ⊦{} ? (□¬□¬¬P↔¬¬□¬□P), i.e.{} ? (□◊¬P↔¬◊□P) 21. S4 ⊦{} ? (¬□¬□¬P↔¬□¬□¬P), i.e.{} ? (¬□◊P↔¬□◊P) or {} ? (¬□◊P↔◊□¬P) or {} ? (◊□¬P↔¬□◊P) or {} ? (◊□¬P↔◊□¬P) 22. S4 ⊦{} ? (□¬□¬□¬P↔¬¬□¬□¬□¬P), i.e. {} ? (□◊□¬P↔¬◊□◊P) or {} ? (□◊□¬P↔¬¬□◊□¬P) or {} ? (□¬□◊P↔¬◊□◊P) or {} ? (□¬□◊P↔¬¬□◊□¬P) 23. S4 ⊦{} ? (¬□¬□¬□¬¬P↔¬□¬□¬□P), i.e.{} ? (◊□◊¬P↔◊□¬□P) or {} ? (◊□◊¬P↔¬□◊□P) or {} ? (¬□◊□¬¬P↔◊□¬□P) or {} ? (¬□◊□¬¬P↔¬□◊□P) Proofs: 11.(a) S4 ⊦{¬□¬P } ? ¬□¬¬□¬P Y { ¬□¬P ; ¬¬□¬P} ? ¬□¬P Y □E {□¬¬□¬P, ¬□¬P } ? ¬¬□¬P {□¬¬□¬P, ¬□¬P } ? ¬□¬P ⊥I {□¬¬□¬P, ¬□¬P } ? ⊥ ¬I {¬□¬P} ? ¬□¬¬□¬P 68 11.(b) S4 ⊦{ ¬□¬¬□¬P } ? ¬□¬P Y Y {□¬P, ¬□¬P } ? □¬P {□¬P, ¬□¬P } ? ¬□¬P ⊥I {□¬P, ¬□¬P } ? ⊥ ¬I { □¬P } ? ¬¬□¬P Y □I {¬□¬¬□¬P, □¬P } ? □¬¬□¬P {¬□¬¬□¬P, □¬P } ? ¬□¬¬□¬P ⊥I {¬□¬¬□¬P, □¬P } ? ⊥ ¬I {¬□¬¬□¬P} ? ¬□¬P 12. (a) S4 ⊦ {□¬□¬□¬□¬P } ? □¬□¬P Y {□¬P; ¬□¬P } ? ¬ □¬P Y □E {□¬P,□¬□¬P } ? □¬P {□¬P; □¬□¬P } ? ¬ □¬P ⊥I {□¬P, □¬□¬P } ? ⊥ Y ¬I {;¬□¬□¬□¬P} ? ¬□¬□¬□¬P {□¬□¬□¬□¬P ,□¬P} ? ¬□¬□¬P □I □E {□¬□¬□¬□¬ P, □¬P } ? ¬□¬□¬□¬P □I {□¬□¬□¬□¬P,□¬P} ? □¬□¬□¬P { □¬P, □¬□¬□¬□¬P } ? ⊥ ¬I {□¬□¬□¬□¬P} ? ¬□¬P □I {□¬□¬□¬□¬P} ? □¬□¬P 69 ⊥I 12. (b) S4 ⊦ {□¬□¬P } ? □¬□¬□¬□¬P Y {□¬□¬P; ¬□¬□¬P } ? ¬□¬□¬P Y □E {□¬□¬P, □¬□¬□¬P } ? ¬□¬□¬P {□¬□¬P, □¬□¬□¬P } ? □¬□¬P ⊥I {□¬□¬P, □¬□¬□¬P} ? ⊥ ¬I {□¬□¬P}? ¬□¬□¬□¬P □I {□¬□¬P}? □¬□¬□¬□¬P 13. (a) S4 ⊦ {¬□¬□¬□¬□P } ? ¬□¬□P Y {□¬□P; ¬□¬□P } ? ¬□¬□P Y □E {□¬□P, □¬□¬□P } ? ¬□¬□P {□¬□P; □¬□¬□P } ? □¬□P ⊥I {□¬□P, □¬□¬□P } ? ⊥ ¬I {□¬□P} ? ¬□¬□¬□P Y □I {¬□¬□¬□¬□P, □¬□P □¬□P} ? □¬□¬□¬□P {¬□¬□¬□¬□P, □¬□P } ? ¬□¬□¬□¬□P ⊥I {¬□¬□¬□¬□P, □¬□P} ? ⊥ ¬I {¬□¬□¬□¬□P} ? ¬□¬□P 70 13. (b) S4 ⊦ {¬□¬□P } ? ¬□¬□¬□¬□P Y {□P; Y ¬□P } ? ¬ □P □E {□P, □¬□P } ? □P {□P, □ ¬□P } ? ¬ □P ⊥I {□P, □¬□P} ? ⊥ Y ¬I {; ¬□¬□¬□P} ? ¬□¬□¬□P {□P} ? ¬□¬□P □I { □P} ? □¬□¬□P □E {□¬□¬□¬□P, □¬P} ? ¬□¬□¬□P ⊥I {□¬□¬□¬□P, □P} ? ⊥ ¬I Y {□¬□¬□¬□P} ? ¬□P □I {¬□¬□P, □¬□¬□¬□P} ? □¬□P {¬□¬□P, □¬□¬□¬□P} ? ¬□¬□P ⊥I {¬□¬□P, □¬□¬□¬□P} ? ⊥ {¬□¬□P} ? ¬□¬□¬□¬□P ¬I Application of the reduction rules: The following is a list of application of reduction rules. The list contains one statement for each of the distinct modalities. The proofs are very long and are not listed here. 24. S4 ⊦ {} ? □P↔□□□P 25. S4 ⊦ {} ? □P↔□□□□P 26. S4 ⊦ {} ? ¬□¬P↔¬□¬¬□¬¬□¬P 27. S4 ⊦ {} ? ¬□¬P↔¬□□□¬P 28. S4 ⊦ {} ? □□¬□¬□¬□¬¬□¬□□P↔□□□□¬□¬¬□¬¬□¬¬□¬□¬□¬□P 29. S4 ⊦ {} ? □□¬□¬□¬□¬¬□¬□□¬□¬P↔□□□□¬□¬¬□¬¬□¬¬□¬□¬□¬□¬□¬P 30. S4 ⊦ {} ? ¬□□¬□¬□¬□¬¬□¬□¬¬□¬P↔¬□□□□¬□¬¬□¬¬□¬¬□¬□¬□¬□¬P 71 31. S4 ⊦ {} ? ¬□□¬□¬□¬□¬¬□P↔¬□□□□¬□¬¬□¬¬□¬¬□¬□¬□P 32. S4 ⊦ {} ? ¬□P↔¬□□□P 33. S4 ⊦ {} ? ¬¬□¬P↔¬¬□¬¬□¬¬□¬P 34. S4 ⊦ {} ? ¬□□¬□¬□¬□¬¬□¬□□P↔¬□□□□¬□¬¬□¬¬□¬¬□¬□¬□¬□P 35. S4 ⊦ {} ? ¬□□¬□¬□¬□¬¬□¬□□¬□¬P↔¬□□□□¬□¬¬□¬¬□¬¬□¬□¬□¬□¬□¬P 36. S4 ⊦ {} ? ¬¬□□¬□¬□¬□¬¬□¬□¬¬□¬P↔¬¬□□□□¬□¬¬□¬¬□¬¬□¬□¬□¬□¬P 37. S4 ⊦ {} ? ¬¬□□¬□¬□¬□¬¬□P↔¬¬□□□□¬□¬¬□¬¬□¬¬□¬□¬□P Some distinct modalities are not equivalent, but one may imply the other. The proofs of these are given in the next pages. S4 ⊦ {} ? (□P → P) (shown above) 38. S4 ⊦ {} ? (P → ¬□¬P) (shown above) 39. S4 ⊦ {} ? (□P → ¬□¬P) i.e. {} ? (□P → ◊P) 40. S4 ⊦ {} ? (□P → ¬□¬□P) i.e. {} ? (□P → ◊□P) 41. S4 ⊦ {} ? (□P → □¬□¬P) i.e. {} ? (□P → □◊P) 42. S4 ⊦ {} ? (□P → ¬□¬□¬□¬P) i.e. {} ? (□P→ ◊□◊P) or {} ? (□P → ¬□◊□¬P) 43. S4 ⊦ {} ? (□P → □¬□¬□P) i.e. {} ? (□P → □◊□P) 44. S4 ⊦ {} ? (¬□¬□P → ¬□¬P) i.e. {} ? (◊□P → ◊P) 45. S4 ⊦ {} ? (□¬□¬P → ¬□¬P) i.e. {} ? (□◊P → ◊P) 46. S4 ⊦ {} ? (¬□¬□¬□¬P → ¬□¬P) i.e. {} ? (◊□◊P → ◊P) or {} ? (¬□◊□¬P → ◊P) 47. S4 ⊦ {} ? (□¬□¬□P → ¬□¬P) i.e. {} ? (□◊□P → ◊P) 48. S4 ⊦ {} ? (¬□¬□P → ¬□¬□¬□¬P) i.e.{}?(◊□P → ◊□◊P) or {} ? (◊□P → ¬□◊□¬P) 49. S4 ⊦ {} ? (□¬□¬□P → ¬□¬□P) i.e. {} ? (□◊□P → ◊□P) 50. S4 ⊦ {} ? (□¬□¬P → ¬□¬□¬□¬P) i.e.{}?(□◊P → ◊□◊P) 72 or {} ? (□◊P → ¬□◊□¬P) 51. S4 ⊦ {} ? (□¬□¬□P → □¬□¬P) i.e. {} ? (□◊□P → □◊P) 52. S4 ⊦ {} ? (□¬□¬□P → ¬□¬□¬□¬P) i.e.{}?(□◊□P→◊□◊P) or {}?(□◊□P→¬□◊□¬P) AProS was made to prove the contrapositives of the above, to verify the interaction between negation and box rules. We do not list the proofs here. 53. S4 ⊦ {} ? (¬P → ¬□P) 54. S4 ⊦ {} ? (¬¬□¬P → ¬P) i.e. {} ? (¬◊P → ¬P) 55. S4 ⊦ {} ? (¬¬□¬P → ¬□P) i.e. {} ? (¬◊P → ¬□P) 56. S4 ⊦ {} ? (¬¬□¬□P → ¬□P) i.e. {} ? (¬◊□P → ¬□P) 57. S4 ⊦ {} ? (¬□¬□¬P → ¬□P) i.e. {} ? (◊□¬P → ¬□P) or {} ? (¬□◊P → ¬□P) 58. S4 ⊦ {} ? (¬¬□¬□¬□¬P → ¬□P) i.e.{} ? (¬◊□¬□¬P → ¬□P) or {} ? (¬¬□◊□¬P → ¬□P) or {} ? (¬¬□¬□◊P → ¬□P)or {} ? (¬◊□◊P → ¬□P) 59. S4 ⊦ {} ? (¬□¬□¬□P → ¬□P) i.e.{} ? (¬□◊□P → ¬□P) or {} ? (◊□¬□P → ¬□P) 60. S4 ⊦ {} ? (¬¬□¬P → ¬¬□¬□P) i.e. {} ? (¬◊P → ¬◊□P) 61. S4 ⊦ {} ? (¬¬□¬P → ¬□¬□¬P) i.e.{} ? (¬◊P → ◊□¬P) or {} ? (¬◊P → ¬□◊P) 62. S4 ⊦ {} ? (¬¬□¬P → ¬¬□¬□¬□¬P) i.e.{} ? (¬◊P → ¬◊□¬□¬P) or {} ? (¬◊P → ¬◊□◊P) or {} ? (¬◊P → ¬¬□◊□¬P)or {} ? (¬◊P → ¬¬□¬□◊P) 63. S4 ⊦ {} ? (¬¬□¬P → ¬□¬□¬□P) or {}? (¬◊P→ ◊□¬□P) or {} ?(¬◊P → ¬□◊□P) 64. S4 ⊦ {} ? (¬¬□¬□¬□¬P → ¬¬□¬□P) i.e. {} ? (¬◊□¬□¬P → ¬◊□P) or {} ? (¬¬□◊□¬P → ¬◊□P)or {} ? (¬¬□¬□◊P → ¬◊□P) or {} ? (¬◊□◊P → ¬◊□P) 65. S4 ⊦ {} ? (¬¬□¬□P → ¬□¬□¬□P) i.e. {} ? (¬◊□P → ◊□¬□P) or {}?(¬◊□P→¬□◊□P) 66. S4 ⊦ {} ? (¬¬□¬□¬□¬P → ¬□¬□¬P) i.e. {} ? (¬◊□¬□¬P → ◊□¬P) or {} ? (¬◊□¬□¬P → ¬□◊P) or {} ? (¬¬□◊□¬P → ◊□¬P) or {} ? (¬¬□◊□¬P → ¬□◊P) or {} ? (¬¬□¬□◊P → ◊□¬P) or {} ? (¬¬□¬□◊P → ¬□◊P) or {} ? (¬◊□◊P → ◊□¬P) or {} ? (¬◊□◊P → ¬□◊P) 73 67. S4 ⊦ {} ? (¬□¬□¬P → ¬□¬□¬□P) i.e.{} ? (◊□¬P → ◊□¬□P) or {} ? (◊□¬P → ¬□◊□P) or {} ? (¬□◊P → ◊□¬□P) or {} ? (¬□◊P → ¬□◊□P) 68. S4 ⊦ {} ? (¬¬□¬□¬□¬P → ¬□¬□¬□P)i.e. {} ? (¬◊□¬□¬P → ◊□¬□P) or {} ? (¬◊□¬□¬P → ¬□◊□P) or {} ? (¬¬□◊□¬P → ◊□¬□P) or {} ? (¬¬□◊□¬P → ¬□◊□P) or {} ? (¬¬□¬□◊P → ◊□¬□P) or {} ? (¬¬□¬□◊P → ¬□◊□P) or {} ? (¬◊□◊P → ◊□¬□P) or {} ? (¬◊□◊P → ¬□◊□P) Proofs. 38. S4 ⊦ {P} ? (¬□¬P) Y {P; ¬P} ? ¬P Y □E {P, □¬ P} ? P {P, □¬P} ? ¬P ⊥I {P,□¬ P} ? ⊥ ¬I {P}? ¬□¬ P 40. S4 ⊦{□P } ? (¬□¬□P) Y {□P; ¬□P} ? ¬□P Y {□P, □¬□P} ? □P {□P, □¬□P} ? ¬□P □E ⊥I {□P, □¬□P} ? ⊥ ¬I {□P}? ¬□¬□ P 74 41. S4 ⊦{□P} ? (□¬□¬P) Y {□P; ¬□P }? ¬□P Y □E {□P, □¬□P }? □P {□P, □¬□P }? ¬□P ⊥I {□P, □¬□P } ? ⊥ ¬I {□P}?¬□¬□P 42. S4 ⊦{□P } ? (¬□¬□¬□¬P) Y Y {□P; ¬P} ? ¬P {□¬ P; P } ? P □E {□P, □¬P} ? P □E {□P, □¬P} ? ¬P ⊥I Y {□P,□¬ P} ? ⊥ ¬I {□P} ? ¬□¬P □I {□P} ? □¬□¬P {□P; ¬□¬□¬P} ? ¬□¬□¬P □E {□P, □¬□¬□¬P} ? ¬□¬□¬P ⊥I {□P,□¬□¬□¬P} ? ⊥ ¬I {□P} ? ¬□¬□¬□¬P 75 43. S4 ⊦{□P } ? (□¬□¬□P) Y {□P; ¬□P }? ¬□P □E Y {□P; □¬□P } ? ¬□P {□P; ¬□P }? □P ⊥I {□P, □¬□P } ? ⊥ ¬I {□P} ? ¬□¬□P □I {□P} ? □¬□¬□P 44. S4 ⊦{¬□¬□P } ? (¬□¬P) Y Y { □¬P; P }? P {□P;¬P }? ¬P □E □E {□¬P ,□P} ? P {□P, □¬P } ? ¬P ⊥I {□¬P ,□P} ? ⊥ ¬I {□¬P} ? ¬□P Y □I {¬□¬□P, □¬P} ? ¬□¬□P {¬□¬□P, □¬P} ? □¬□P ⊥I {¬□¬□P, □¬P} ? ⊥ ¬I {¬□¬□P} ? ¬□¬P 76 45. S4 ⊦{□¬□¬P } ? (¬□¬P) Y {;¬□¬P} ? ¬□¬P □E {□¬□¬P} ? ¬□¬P 46. S4 ⊦{¬□¬□¬□¬P } ? (¬□¬P) Y {□¬P; ¬□¬P } ? ¬□¬P Y □E {□¬P, □¬□¬P } ? ¬□¬P {□¬P; ¬□¬P } ? □¬P ⊥I {□¬P, □¬□¬P } ? ⊥ ¬I {□¬P} ? ¬□¬□¬P Y □I {¬□¬□¬□¬P,□¬P} ? □¬□¬□¬P {¬□¬□¬□¬P, □¬P } ? ¬□¬□¬□¬P ⊥I □¬□P {¬□¬□¬□¬P, □¬P} ? ⊥ ¬I {¬□¬□¬□¬P} ? ¬□¬P 77 47. S4 ⊦ {□¬□¬□P } ? (¬□¬P) Y Y {□P; ¬P } ? ¬P {□¬P; P} ? P □E □E {□¬P, □P} ? ¬P {□¬P, □P} ? P ⊥I {□¬P, □P} ? ⊥ Y ¬I {□¬P; ¬□¬□P } ? ¬□¬□P {□¬P} ? ¬□P □E □I {□¬□¬□P, □¬P} ? ¬□¬□P {¬□¬□P, □¬P} ? □¬□P ⊥I {□¬□¬□P, □¬P} ? ⊥ ¬I {□¬□¬□P} ? ¬□¬P 48. S4 ⊦{¬□¬□P } ? (¬□¬□¬□¬P) Y Y {□P; ¬P } ? ¬P {□¬P; P} ? P { □P, □¬P} ? P □E □E {□¬P, □P } ? ¬P ⊥I {□P, □¬P} ? ⊥ ¬I {□P} ? ¬□¬P Y □I {□¬□¬□¬P, □P} ? □¬□¬P {□¬□¬□¬P, □¬P} ? ¬□¬□¬P ⊥I {□¬□¬□¬P, □P} ? ⊥ ¬I {□¬□¬□¬P} ? ¬□P Y □I {¬□¬□P, □¬□¬□¬P} ? □¬□P {¬□¬□P, □¬□¬□¬P} ? ¬□¬□P ⊥I {¬□¬□P, □¬□¬□¬P} ? ⊥ ¬I {¬□¬□P} ? ¬□¬□¬□¬P 78 49. S4 ⊦{□¬□¬□P } ? (¬□¬□P) Y {; ¬□¬□P} ? ¬□¬□P □E {□¬□¬□P} ? ¬□¬□P 50. S4 ⊦{□¬□¬P } ? (¬□¬□¬□¬P) Y {□¬□¬P; ¬□¬□¬P } ? ¬□¬□¬P Y □¬□¬P¬□¬P □E {□¬□¬P, □¬□¬□¬P } ? ¬□¬□¬P {□¬□¬P, ¬□¬□¬P } ? □¬□¬P ⊥I □¬□¬P¬□¬P {□¬□¬P, □¬□¬□¬P } ? ⊥ ¬I {□¬□¬P} ? ¬□¬□¬□¬P 79 51. S4 ⊦ {□¬□¬□P } ? (□¬□¬P) Y Y {□¬P; P } ? P {P; □¬P } ? ¬P □E □E {□P, □¬P } ? P {□P, □¬P } ? ¬ P ⊥I {□¬P, □P} ? ⊥ ¬I {□¬P} ? ¬□P Y □I {¬□¬□P, □¬P} ? □¬□P {¬□¬□P, □¬P} ? ¬□¬□P ⊥I {¬□¬□P, □¬P} ? ⊥ ¬I {¬□¬□P} ? ¬□¬P □I {□¬□¬□P} ? □¬□¬P 80 52. S4 ⊦{□¬□¬□P } ? (¬□¬□¬□¬P) Y Y {□P; ¬P }? ¬P {□¬P; P }? P □E □E {□P, □ ¬P }? ¬P {□P, □ ¬P }? P ⊥I {□¬P, □P} ? ⊥ ¬I Y {□¬P}?¬□P {□¬P; ¬□¬□P } ? ¬□¬□P □I {□¬□¬□P, □¬P} ? □¬□P □E {□¬□¬□P, □¬P} ? ¬□¬□P ⊥I Y {□¬□¬□P, □¬P} ? ⊥ ¬I {□¬□¬□P; {□¬□¬□P} ? ¬□¬P ¬□¬□¬P } ? ¬□¬□¬P □E □I {□¬□¬□P, □¬□¬□¬P } ? □¬□¬P {□¬□¬□P,□¬□¬□¬P } ? ¬□¬□¬P ⊥I {□¬□¬□P, □¬□¬□¬P } ? ⊥ ¬I {□¬□¬□P} ? ¬□¬□¬□¬P Proofs of a few theorems that are propositional variants of the reduction rules. 69. S4 ⊦{¬□¬P} ? (¬□□¬P) Y {¬□¬P; □¬P } ? □¬P Y □E {¬□¬P; □□¬P } ? □¬P {¬□¬P; □¬P } ? ¬□¬P ⊥I {¬□¬P, □□¬P } ? ⊥ ¬I {¬□¬P}?¬□□¬P 81 70. S4 ⊦ {¬□ 43. □¬P} S4 ⊦ {¬□ ? (¬□¬P □¬P}) ? (¬□¬P ) Y { □¬P } ? □¬P Y □I {¬□□¬P, □¬P } ? ¬□□¬P {¬□□¬P, □¬P } ? □□¬P ⊥I {¬□□¬P, □¬P } ? ⊥ ¬I {¬□□¬P}?¬□¬P {¬□□¬P} ? ¬□¬P IV Miscellaneous These examples show the interaction of box and negation rules with the other connectives. 71. S4 ⊦ {R, □S} ? □(R→S), Y {R; S} ? S I {;S} ? (RS) □E {□S} ? (RS) □I {R, □S} ? □(RS) 72. S4 ⊦ {□R, □S} ? □(RVS) Y {□R; S}? (RVS) □E {□R, □S}? (RVS) □I {□R, □S}? □(RVS) 82 Modification of distributivity 73. S4 ⊦{□(B&C)} ? (□B&□C) Y Y {;(B&C), B}? B {;(B&C), C}? C &E &E {;(B&C)}? B {;(B&C)}? C □E □E ) {□(B&C)}? B {□(B&C)}? C □I □I {)□(B&C)}? □B {□(B&C)}? □C &I {□(B&C)}? □B&□C 74. S4 ⊦ {□P} ? □¬□¬P Y Y {□¬P; P} ? P {□P; ¬P} ? ¬P □E □E {□¬P, □P } ? P {□P, □¬P} ? ¬P ⊥I {□P,□¬ P} ? ⊥ ¬I {□P}? ¬□¬ P □I {□P}? □¬□¬ P 83 75. S4 ⊦{□¬P} ? ¬□P Y Y {□¬P; P }? P {□P; ¬P }? ¬P □E □E {□P, □ ¬P }? P {□P, □ ¬P }? ¬P ⊥I {□¬P ,□P} ? ⊥ ¬I {□¬P}?¬□P 76. S4 ⊦ {□(R→S), □R, R} ? □S Y Y {;R→S,R,S}? S {;R→S,R}? R E {;R→S,R}? S □E {□(R→S);R}?S □E □(AA) {□(R→S),□R}?S □I (AA) {□□(R →S),□R,R}?□S □(AA) 77. S4 ⊦ {□P, ¬P} ? Q Y { ¬P, ¬□Q ; P }? P Y □E { ¬P, ¬□Q, □P }? P { ¬P, ¬□Q ; P }? ¬P ⊥I □(AA) {□P, ¬P, ¬□Q }? ⊥ ¬E {□P, ¬P} ? □Q □(AA) 84 The following propositional variants were also proved. 78. S4 ⊦{A, □(B→C), (A→□B)} ? □C 79. S4 ⊦ {□P} ? □P, 80. S4 ⊦ {□R & □S} ? □R, 81. S4 ⊦{□R, □S} ? □(R→S), 82. S4 ⊦ {□R, □S} ? □□(R→S), 83. S4 ⊦{□R, □S} ? □(R&S), 84. S4 ⊦ {□((R↔S)&Q), □(RVZ),□¬Z} ? □(SVP) 85. S4 ⊦{¬□A→B, ¬□A→¬B} ? □A 86. S4 ⊦ {¬□A→B, ¬□A→¬B, □(A→C)} ? □C 87. S4 88. S4 ⊦ {□P, ¬P} ? Q, 89. S4 ⊦{□R, □S} ? □(R&S), 90. S4 ⊦ □(□P→P) 91. S4 ⊦ P→□(□P→P) ⊦ {P} ? ¬□¬P 85 Unprovable statements: The distinct modalities are unboxed, □, ◊, ◊□, □◊, ◊□◊, □◊□ and their negations P is distinct from other modalities 1. S4 ⊬ 2. S4 ⊬ {} ? (P ↔ ¬□¬P), 3. S4 ⊬ {} ? (P ↔ ¬□¬□P), i.e.{} ? (P ↔ ◊□P) 4. S4 ⊬ {} ? (P ↔ □¬□¬P), i.e.{} ? (P ↔ □◊P) 5. S4 ⊬ {} ? (P ↔ ¬□¬□¬□¬P), i.e. {} ? (P ↔ ◊□¬□¬P) or {} ? (P ↔ ¬□◊□¬P) or {} ? (P ↔ □P), i.e.{} ? (P ↔ ◊P), {} ? (P ↔ ¬□¬□◊P) or {} ? (P ↔ ◊□◊P) 6. S4 ⊬ {} ? (P ↔ □¬□¬□P), i.e. {} ? (P ↔ □◊□P) 7. S4 ⊬ {} ? (P ↔ ¬P), 8. S4 ⊬ {} ? (P ↔ ¬□P), 9. S4 ⊬ {} ? (P ↔ ¬¬□¬P), i.e.{} ? (P ↔ ¬◊P) 10. S4 ⊬ {} ? (P ↔ ¬¬□¬□P), i.e. {} ? (P ↔ ¬◊□P) 11. S4 ⊬ {} ? (P ↔ ¬□¬□¬P), i.e. {} ? (P ↔ ¬□◊P) or {} ? (P ↔ ◊□¬P) 12. S4 ⊬ {} ? (P ↔ ¬¬□¬□¬□¬P), i.e. {} ? (P ↔ ¬◊□¬□¬P) or {} ? (P ↔ ¬¬□◊□¬P) or {} ? (P ↔ ¬¬□¬□◊P) or {} ? (P ↔ ¬◊□◊P) 13. S4 ⊬ {} ? (P ↔ ¬□¬□¬□P), i.e. {} ? (P ↔ ◊□¬□P) or {} ? (P ↔ ¬□◊□P) □P is distinct from other modalities 14. S4 ⊬ {} ? (□P ↔ ¬□¬P), i.e.{} ? (□P ↔ ◊P) 15. S4 ⊬ {} ? (□P ↔ ¬□¬□P), i.e. {} ? (□P ↔ ◊□P) 16. S4 ⊬ 17. S4 ⊬ {} ? (□P ↔ ¬□¬□¬□¬P), i.e. {} ? (□P ↔ ◊□¬□¬P) or {} ? (□P ↔ ¬□◊□¬P) or {} ? (□P ↔ □¬□¬P), i.e. {} ? (□P ↔ □◊P) {} ? (□P ↔ ¬□¬□◊P) or {} ? (□P ↔ ◊□◊P) 86 18. S4 ⊬ {} ? (□P ↔ □¬□¬□P), i.e. {} ? (□P ↔ □◊□P) 19. S4 ⊬ {} ? (□P ↔ ¬P) 20. S4 ⊬ {} ? (□P ↔ ¬□P) 21. S4 ⊬ {} ? (□P ↔ ¬¬□¬P), i.e.{} ? (□P ↔ ¬◊P) 22. S4 ⊬ {} ? (□P ↔ ¬¬□¬□P), i.e.{} ? (□P ↔ ¬¬◊□P) 23. S4 ⊬ {} ? (□P ↔ ¬□¬□¬P), i.e. {} ? (□P ↔ ¬□◊P) or {} ? (□P ↔ ◊□¬P) 24. S4 ⊬ {} ? (□P ↔ ¬¬□¬□¬□¬P), i.e.{} ? (□P ↔¬◊□¬□¬P) or {} ? (□P ↔ ¬¬□◊□¬P) or {} ? (□P ↔ ¬¬□¬□◊P) or {} ? (□P ↔ ¬◊□◊P) 25. S4 ⊬ {} ? (□P ↔ ¬□¬□¬□P), i.e.{} ? (□P ↔ ◊□¬□P) or {} ? (□P ↔ ¬□◊□P) ◊P is distinct from other modalities 26. S4 ⊬ {} ? (¬□¬P ↔ ¬□¬□P), i.e.{} ? (◊P ↔ ◊□P) 27. S4 ⊬ {} ? (¬□¬P ↔ □¬□¬P), 28. S4 ⊬ {} ? (¬□¬ P↔ ¬□¬□¬□¬P), i.e. {} ? (◊P ↔ ◊□¬□¬P) or {} ? (◊P ↔ ¬□◊□¬P) i.e.{} ? (◊P ↔ □◊P) or {} ? (◊P ↔ ¬□¬□◊P) or {} ? (◊P ↔ ◊□◊P) 29. S4 ⊬ {} ? (¬□¬P ↔ □¬□¬□P), i.e.{} ? (◊P ↔ □◊□P) 30. S4 ⊬ {} ? (¬□¬P ↔ ¬P), 31. S4 ⊬ {} ? (¬□¬P ↔ ¬□P), i.e.{} ? (◊P ↔ ¬□P) 32. S4 ⊬ 33. S4 ⊬ {} ? (¬□¬P ↔ ¬¬□¬□P), i.e.{} ? (◊P ↔ ¬◊□P) 34. S4 ⊬ {} ? (¬□¬P ↔ ¬□¬□¬P), i.e.{} ? (◊P ↔ ¬□◊P) or {} ? (◊P ↔ ◊□¬P) 35. S4 ⊬ {} ? (¬□¬P ↔ ¬¬□¬□¬□¬P), i.e.{} ?( ◊P ↔¬◊□¬□¬P) or {} ?( ◊P↔¬¬□◊□¬P) i.e.{} ? (◊P ↔ ¬P) {} ? (¬□¬P ↔ ¬¬□¬P), i.e.{} ? (◊P ↔ ¬◊P) or {} ? (◊P ↔ ¬¬□¬□◊P) or {} ? (◊P ↔ ¬◊□◊P) 36. S4 ⊬ {} ? (¬□¬P ↔ ¬□¬□¬□P), i.e.{} ? (◊P ↔ ◊□¬□P) or {} ? (◊P ↔ ¬□◊□P) □◊P is distinct from other modalities 37. S4 ⊬ {} ?(¬□¬□P ↔ □¬□¬P), i.e.{} ?( ◊□P ↔ □◊P) 87 38. S4 ⊬ {} ? (¬□¬□P ↔ ¬□¬□¬□¬P), i.e.{}?( ◊□P ↔ ◊□¬□¬P) or {}?( ◊□P↔¬□◊□¬P) or {}?(◊□P ↔ ¬□¬□◊P) or {}? (◊□P↔ ◊□◊P) 39. S4 ⊬ {} ?(¬□¬□P ↔ □¬□¬□P), i.e.{} ?( ◊□P ↔ □◊□P) 40. S4 ⊬ {} ? (¬□¬□P ↔ ¬P), 41. S4 ⊬ {} ? (¬□¬□P ↔ ¬□P), i.e.{} ? (◊□P ↔ ¬□P) 42. S4 ⊬ {} ? (¬□¬□P ↔ ¬¬□¬P), i.e.{} ? (◊□P ↔ ¬◊P) 43. S4 ⊬ {} (¬□¬□P ↔ ¬¬□¬□P), i.e.{} (◊□P ↔ ¬◊□P) 44. S4 ⊬ {} ? (¬□¬□P ↔ ¬□¬□¬P), i.e.{} ? (◊□P ↔ ¬□◊P) or {} ? (◊□P ↔ ◊□¬P) 45. S4 ⊬ {} ?(¬□¬□P ↔ ¬¬□¬□¬□¬P), i.e.{}?( ◊□P↔¬◊□¬□¬P) or {}?(◊□P↔¬¬□◊□¬P) i.e.{} ? (◊□P ↔ ¬P) or {}?(◊□P ↔ ¬¬□¬□◊P) or {}?(◊□P ↔ ¬◊□◊P) 46. S4 ⊬ {} ? (¬□¬□P ↔ ¬□¬□¬□P), i.e.{} ? (◊□P ↔ ◊□¬□P) or {} ? (◊□P ↔ ¬□◊□P) □◊P is distinct from other modalities 47. S4 ⊬ {} ? (□¬□¬P ↔ ¬□¬□¬□¬P), i.e.{}?( □◊P ↔ ◊□¬□¬P) or {}?(□◊P↔¬□◊□¬P) or {}?(□◊P ↔ ¬□¬□◊P) or {}? (□◊P↔ ◊□◊P) 48. S4 ⊬ {} ? (□¬□¬P ↔ □¬□¬□P), i.e.{} ? (□◊P ↔ □◊□P) 49. S4 ⊬ {} ? (□¬□¬P ↔ ¬P), i.e.{} ? (□◊P ↔ ¬P) 50. S4 ⊬ {} ? (□¬□¬P ↔ ¬□P), i.e.{} ? (□◊P ↔ ¬□P) 51. S4 ⊬ {} ? (□¬□¬P ↔ ¬¬□¬P), i.e.{} ? (□◊P ↔ ¬◊P) 52. S4 ⊬ {} ? (□¬□¬P ↔ ¬¬□¬□P), i.e.{} ? (□◊P ↔ ¬◊□P) 53. S4 ⊬ {} ? (□¬□¬P ↔ ¬□¬□¬P), i.e.{} ? (□◊P ↔ ¬□◊P) or {} ? (□◊P ↔ ◊□¬P) 54. S4 ⊬ {} ? (□¬□¬P ↔ ¬¬□¬□¬□¬P), i.e.{}?(□◊P↔¬◊□¬□¬P) or {}?(□◊P↔¬¬□◊□¬P) or {}?(□◊P ↔ ¬¬□¬□◊P) or {}?(□◊P ↔ ¬◊□◊P) 55. S4 ⊬ {} ? (□¬□¬P ↔ ¬□¬□¬□P), i.e.{} ? (□◊P ↔ ◊□¬□P) or {} ? (□◊P ↔ ¬□◊□P) ◊□◊P is distinct from other modalities 56. S4 ⊬ {} ? (¬□¬□¬□¬P ↔ □¬□¬□P), i.e.{}?(◊□◊P ↔ □◊□P) or {}?(◊□¬□¬P ↔ □◊□P) or {} ? (¬□¬□◊P ↔ □◊□P) or {} ? (¬□◊□¬P ↔ □◊□P) 88 57. S4 ⊬ {} ? (¬□¬□¬□¬P ↔ ¬P), i.e.{}?(◊□◊P ↔ ¬P) or {}?(◊□¬□¬P ↔ ¬P) or {} ? (¬□¬□◊P ↔ ¬P) or {} ? (¬□◊□¬P ↔ ¬P) 58. S4 ⊬ {} ? (¬□¬□¬□¬P ↔ ¬□P), i.e.{}?(◊□◊P ↔ ¬□P) or {}?(◊□¬□¬P ↔ ¬□P) or {} ? (¬□¬□◊P ↔ ¬□P) or {} ? (¬□◊□¬P ↔ ¬□P) 59. S4 ⊬ {} ? (¬□¬□¬□¬P ↔ ¬¬□¬P), i.e.{}?(◊□◊P ↔ ¬◊P) or {}?(◊□¬□¬P ↔ ¬◊P) or {} ? (¬□¬□◊P ↔ ¬◊P) or {} ? (¬□◊□¬P ↔ ¬◊P) 60. S4 ⊬ {} ? (¬□¬□¬□¬P ↔ ¬¬□¬□P), i.e.{}?(◊□◊P ↔ ¬◊□P) or {}?(◊□¬□¬P ↔ ¬◊□P) or {} ? (¬□¬□◊P ↔ ¬◊□P) or {} ? (¬□◊□¬P ↔ ¬◊□P) 61. S4 ⊬ {} ? (¬□¬□¬□¬P ↔ ¬□¬□¬P), i.e.{}?(◊□◊P ↔ ¬◊□P) or {}?(◊□¬□¬P ↔ ¬◊□P) or {} ? (¬□¬□◊P ↔ ¬◊□P) or {} ? (¬□◊□¬P ↔ ¬◊□P) 62. S4 ⊬ {} ? (¬□¬□¬□¬P ↔ ¬¬□¬□¬□¬P), i.e.{}?(◊□◊P ↔ ¬◊□¬□¬P) or {}?(◊□¬□¬P↔ ¬◊□¬□¬P) or {} ? (¬□¬□◊P ↔ ¬◊□¬□¬P) or {} ? (¬□◊□¬P ↔ ¬◊□¬□¬P) or {}?(◊□◊P ↔ □◊□P) or {}?(◊□¬□¬P↔ ¬¬□◊□¬P) or {} ? (¬□¬□◊P ↔ ¬¬□◊□¬P) or {} ? (¬□◊□¬P ↔ ¬¬□◊□¬P) or {}?(◊□◊P ↔ □◊□P) or {}?(◊□¬□¬P↔ ¬¬□¬□◊P) or {} ? (¬□¬□◊P ↔ ¬¬□¬□◊P) or {} ? (¬□◊□¬P ↔ ¬¬□¬□◊P) or {}?(◊□◊P ↔ □◊□P) or {}?(◊□¬□¬P↔ ¬◊□◊P) or {} ? (¬□¬□◊P ↔ ¬◊□◊P) or {} ? (¬□◊□¬P ↔ ¬◊□◊P) 63. S4 ⊬ {} ? (¬□¬□¬□¬P ↔ ¬□¬□¬□P), i.e.{}?(◊□◊P ↔ ◊□¬□P) or {}? (◊□¬□¬P ↔ ◊□¬□P) or {} ? (¬□¬□◊P ↔ ◊□¬□P) or {} ? (¬□◊□¬P ↔ ◊□¬□P) or {}?(◊□◊P ↔ ¬□◊□P) or {}? (◊□¬□¬P ↔ ¬□◊□P) or {} ? (¬□¬□◊P ↔ ¬□◊□P) or {} ? (¬□◊□¬P ↔ ◊□¬□P) 64. S4 ⊬ 65. S4 ⊬ {} ? (¬P ↔ ¬¬□¬P), i.e.{} ? (¬P ↔ ¬◊P) 66. S4 ⊬ {} ? (¬P ↔ ¬¬□¬□P), i.e.{} ? (¬P ↔ ¬◊□P) 67. S4 ⊬ {} ? (¬P ↔ ¬□¬□¬P), i.e.{} ? (¬P ↔ ¬□◊P) or {} ? (¬P ↔ ◊□¬P) 68. S4 ⊬ {} ? (¬P ↔ ¬¬□¬□¬□¬P), i.e.{}?(¬P↔¬◊□¬□¬P) or {}?(¬P↔¬¬□◊□¬P) {} ? (¬P ↔ ¬□P), i.e.{} ? (¬P ↔ ¬□P) or {}?(¬P ↔ ¬¬□¬□◊P) or {}?(¬P ↔ ¬◊□◊P) 69. S4 ⊬ {} ? (¬P ↔ ¬□¬□¬□P), i.e.{} ? (¬P ↔ ◊□¬□P) or {} ? (¬P ↔ ¬□◊□P) ¬□P is distinct from other modalities 70. S4 ⊬ {} ? (¬□P ↔ ¬¬□¬P), i.e.{} ? (¬□P ↔ ¬◊P) 71. S4 ⊬ {} ? (¬□P ↔ ¬¬□¬□P), i.e.{} ? (¬□P ↔ ¬◊□P) 72. S4 ⊬ {} ? (¬□P ↔ ¬□¬□¬P), i.e.{} ? (¬□P ↔ ¬□◊P) or {} ? (¬□P ↔ ◊□¬P) 89 73. S4 ⊬ {} ? (¬□P ↔ ¬¬□¬□¬□¬P), i.e.{}?(¬□P↔¬◊□¬□¬P) or {}?(¬□P↔¬¬□◊□¬P) or {}?(¬□P ↔ ¬¬□¬□◊P) or {}?(¬□P ↔ ¬◊□◊P) 74. S4 ⊬ {} ? (¬□P ↔ ¬□¬□¬□P), i.e.{} ? (¬□P ↔ ◊□¬□P) or {} ? (¬□P ↔ ¬□◊□P) 75. S4 ⊬ {} ? (¬¬□¬P ↔ ¬¬□¬□P), i.e.{} ? (¬◊P ↔ ¬◊□P) 76. S4 ⊬ {} ? (¬¬□¬P ↔ ¬□¬□¬P), i.e.{} ? (¬◊P ↔ ¬□◊P) or {} ? (¬◊P ↔ ◊□¬P) 77. S4 ⊬ {} (¬¬□¬P ↔ ¬¬□¬□¬□¬P), i.e.{}?(¬◊P↔¬◊□¬□¬P) or {}?(¬◊P↔¬¬□◊□¬P) or {}?(¬◊P ↔ ¬¬□¬□◊P) or {}?(¬◊P ↔ ¬◊□◊P) 78. S4 ⊬ {} ? (¬¬□¬P ↔ ¬□¬□¬□P), i.e.{} ? (¬◊P ↔ ◊□¬□P) or {} ? (¬◊P ↔ ¬□◊□P) ¬◊□P is distinct from other modalities 79. S4 ⊬ {} ? (¬¬□¬□P ↔ ¬□¬□¬P), i.e.{} ? (¬◊□P ↔ ¬□◊P) or {} ? (¬◊□P ↔ ◊□¬P) 80. S4 ⊬ {} ? (¬¬□¬□P ↔ ¬¬□¬□¬□¬P), i.e.{}?(¬◊□P↔¬◊□¬□¬P) or {}?(¬◊□P↔¬¬□◊□¬P) or {}?(¬◊□P ↔ ¬¬□¬□◊P) or {}?(¬◊□P ↔ ¬◊□◊P) 81. S4 ⊬ {} ? (¬¬□¬□P ↔ ¬□¬□¬□P), i.e.{} ? (¬◊□P ↔ ◊□¬□P) or {} ? (¬◊□P ↔¬□◊□P) 82. S4 ⊬ {} ? (¬□¬□¬P ↔ ¬¬□¬□¬□¬P), i.e.{}?(◊□¬P↔¬◊□¬□¬P) or {}?(◊□¬P↔¬¬□◊□¬P) or {}?(◊□¬P ↔ ¬¬□¬□◊P) or {}?(◊□¬P ↔ ¬◊□◊P) or {}?(¬◊□P↔¬◊□¬□¬P) or {}?(¬□◊P↔¬¬□◊□¬P) or {}?(¬□◊P ↔ ¬¬□¬□◊P) or {}?(¬□◊P ↔ ¬◊□◊P) 83. S4 ⊬ {} ? (¬□¬□¬P ↔ ¬□¬□¬□P), i.e.{} ? (◊□¬P ↔ ◊□¬□P) or {} ? (◊□¬P ↔¬□◊□P) or {} ? (¬□◊P ↔ ◊□¬□P) or {} ? (¬□◊P ↔¬□◊□P) 84. S4 ⊬ {} ?(¬¬□¬□¬□¬P ↔ ¬□¬□¬□P)}, i.e. or {}?(¬◊□◊P ↔ ◊□¬□P) or {}? (¬◊□¬□¬P ↔ ◊□¬□P) or {} ? (¬¬□¬□◊P ↔ ◊□¬□P) or {} ? (¬¬□◊□¬P ↔ ◊□¬□P) or {}?(¬◊□◊P ↔ ¬□◊□P) or {}? (¬◊□¬□¬P ↔ ¬□◊□P) or {} ? (¬¬□¬□◊P ↔ ¬□◊□P) or {} ? (¬¬□◊□¬P ↔ ◊□¬□P) The following modalitles are totally distinct. We have shown above that they are not equivalent. So, we only need to show one side of this. 85. S4 ⊬ {} ? (P → ¬□¬□P), i.e.{} ? (P → ◊□P) 86. S4 ⊬ {} ? (P → □¬□¬P), i.e.{} ? (P → □◊P) 87. S4 ⊬ {} ? (P → ¬□¬□¬□¬P), i.e. {} ? (P → ◊□¬□¬P) or {} ? (P → ¬□◊□¬P) or {} ? (P → ¬□¬□◊P) or {} ? (P → ◊□◊P) 88. S4 ⊬ {} ? (P → □¬□¬□P), i.e.{} ? (P → □◊□P) 90 89. S4 ⊬ {} ? (¬□¬□P → □¬□¬P), i.e.{} ? (◊□P → □◊P) 90. S4 ⊬ 91. S4 ⊬ {} ? (¬¬□¬□P → □¬□¬P), i.e.{} ? (¬◊□P → □◊P) 92. S4 ⊬ {} ? (¬¬□¬□P → □¬□¬P), i.e.{} ? (¬◊□P → □◊P) 93. S4 ⊬ {} ? (P → ¬¬□¬□P), i.e.{} ? (P → ¬◊□P) 94. S4 ⊬ {} ? (P → ¬□¬□¬P), i.e.{} ? (P → ◊□¬P) or {} ? (P → ¬□◊P) 95. S4 ⊬ {} ? (P → ¬¬□¬□¬□¬P), i.e.{} ? (P → ¬◊□¬□¬P) or {} ? (P → ¬¬□◊□¬P) or {} ? (¬□¬□P → □¬□¬P), i.e.{} ? (◊□P → □◊P) {} ? (P → ¬¬□¬□◊P) or {} ? (P → ¬◊□◊P) 96. S4 ⊬ {} ? (P → ¬□¬□¬□P), i.e.{} ? (P → ◊□¬□P) or {} ? (P → ¬□◊□P) 97. S4 ⊬ {} ? (¬□¬□P → ¬□¬□¬P), i.e.{} ? (◊□P → ◊□¬P) or {} ? (◊□P → ¬□◊P) 98. S4 ⊬ {} ? (¬P → ¬¬□¬□P), i.e.{} ? (¬P → ¬◊□P) 99. S4 ⊬ {} ? (¬P → ¬□¬□¬P), i.e.{} ? (¬P → ◊□¬P) or {} ? (¬P → ¬□◊P) 100. S4 ⊬ {} ? (¬P → ¬¬□¬□¬□¬P), i.e.{} ? (¬P → ¬◊□¬□¬P) or {} ? (¬P → ¬¬□◊□¬P) or {} ? (¬P → ¬¬□¬□◊P) or {} ? (¬P → ¬◊□◊P) 101. S4 ⊬ {} ? (¬P → ¬□¬□¬□P), i.e.{} ? (¬P → ◊□¬□P) or {} ? (¬P → ¬□◊□P) 102. S4 ⊬ {} ? (¬¬□¬□P → ¬□¬□¬P), i.e.{} ? (¬◊□P → ◊□¬P) or {} ? (¬◊□P → ¬□◊P) 103. S4 ⊬ {} ? (¬¬□¬□P → ¬□¬□¬P), i.e.{} ? (¬◊□P → ◊□¬P) or {} ? (¬◊□P → ¬□◊P) 104. S4 ⊬ {} ? (P → ¬P), 105. S4 ⊬ {} ? (P → ¬□P), 106. S4 ⊬ {} ? (P → ¬¬□¬P), i.e.{} ? (P → ¬◊P) 107. S4 ⊬ {} ? (¬P → ¬¬□¬□P), i.e.{} ? (¬P → ¬◊□P) 108. S4 ⊬ {} ? (¬P → ¬□¬□¬P), i.e.{} ? (¬P → ◊□¬P) or {} ? (¬P → ¬□◊P) 109. S4 ⊬ {} ? (¬P → ¬¬□¬□¬□¬P), i.e.{} ? (¬P → ¬◊□¬□¬P) or {} ? (¬P → ¬¬□◊□¬P) or {} ? (¬P → ¬¬□¬□◊P) or {} ? (¬P → ¬◊□◊P) 91 110. S4 ⊬ {} ? (¬P → ¬□¬□¬□P), i.e.{} ? (¬P → ◊□¬□P) or {} ? (¬P → ¬□◊□P) 111. S4 ⊬ {} ? (□P → ¬P), 112. S4 ⊬ {} ? (□P → ¬□P), 113. S4 ⊬ {} ? (□P → ¬¬□¬P), i.e.{} ? (□P → ¬◊P) 114. S4 ⊬ {} ? (□P → ¬¬□¬□P), i.e.{} ? (□P → ¬◊□P) 115. S4 ⊬ {} ? (□P → ¬□¬□¬P), i.e.{} ? (□P → ◊□¬P) or {} ? (□P → ¬□◊P) 116. S4 ⊬ {} ? (□P → ¬¬□¬□¬□¬P), i.e.{} ? (□P → ¬◊□¬□¬P) or {} ? (□P → ¬¬□◊□¬P) or {} ? (□P → ¬¬□¬□◊P) or {} ? (□P → ¬◊□◊P) 117. S4 ⊬ {} ? (□P → ¬□¬□¬□P), i.e.{} ? (□P → ◊□¬□P) or {} ? (□P → ¬□◊□P) 118. S4 ⊬ {} ? (¬□¬P → ¬P), i.e.{} ? (◊P → ¬P) 119. S4 ⊬ {} ? (¬□¬P → ¬□P), i.e.{} ? (◊P → ¬□P) 120. S4 ⊬ {} ? (¬□¬P → ¬¬□¬P), i.e.{} ? (◊P → ¬◊P) 121. S4 ⊬ {} ? (¬□¬P → ¬¬□¬□P), i.e.{} ? (◊P → ¬◊□P) 122. S4 ⊬ {} ? (¬□¬P → ¬□¬□¬P), i.e.{} ? (◊P → ◊□¬P) or {} ? (◊P → ¬□◊P) 123. S4 ⊬ {} ? (¬□¬P → ¬¬□¬□¬□¬P), i.e.{} ? (◊P → ¬◊□¬□¬P) or {} ? (◊P → ¬¬□◊□¬P) or {} ? (◊P → ¬¬□¬□◊P) or {} ? (◊P → ¬◊□◊P) 124. S4 ⊬ {} ? (¬□¬P → ¬□¬□¬□P), i.e.{} ? (◊P → ◊□¬□P) or {} ? (◊P → ¬□◊□P) 125. S4 ⊬ {} ? (¬□¬□P → ¬P), i.e.{} ? (◊□P → ¬P) 126. S4 ⊬ {} ? (¬□¬□P→¬□P), i.e.{} ? (◊□P→¬□P) 127. S4 ⊬ {} ? (¬□¬□P → ¬¬□¬P), i.e.{} ? (◊□P → ¬◊P) 128. S4 ⊬ {} ?(¬□¬□P → ¬¬□¬□P), i.e.{} ?( ◊□P → ¬◊□P) 129. S4 ⊬ {} ? (¬□¬□P → ¬□¬□¬P), i.e.{} ? (◊□P → ◊□¬P) or {} ? (◊□P → ¬□◊P) 130. S4 ⊬ {} ? (¬□¬□P → ¬¬□¬□¬□¬P), i.e.{} ? (◊□P → ¬◊□¬□¬P) {} ? (◊□P → ¬¬□¬□◊P) or {} ? (◊□P → ¬◊□◊P) 92 or {} ? (◊□P → ¬¬□◊□¬P) or 131. S4 ⊬ {} ? ( ¬□¬□P → ¬□¬□¬□P), i.e.{} ? (◊□P → ◊□¬□P) or {} ? (◊□P → ¬□◊□P) 132. S4 ⊬ {} 133. S4 ⊬ {} ? (□¬□¬P → ¬□P), i.e.{} ? (□◊P → ¬□P) 134. S4 ⊬ {} ? (□¬□¬P → ¬¬□¬P), i.e.{} ? (□◊P → ¬◊P) 135. S4 ⊬ {} ? (□¬□¬P → ¬¬□¬□P), i.e.{} ? (□◊P → ¬◊□P) 136. S4 ⊬ {} ? (□¬□¬P → ¬□¬□¬P), i.e.{} ? (□◊P → ¬□◊P) or {} ? (□◊P → ◊□¬P) 137. S4 ⊬ {} ? (□¬□¬P → ¬¬□¬□¬□¬P), i.e.{} ? (□◊P → ¬◊□¬□¬P) ? (□¬□¬P → ¬P), i.e.{} ? (□◊P → ¬P) or {} ? (□◊P → ¬¬□◊□¬P) or {} ? (□◊P → ¬¬□¬□◊P) or {} ? (□◊P → ¬◊□◊P) 138. S4 ⊬ {} ? (□¬□¬P → ¬□¬□¬□P), i.e.{} ? (□◊P → ◊□¬□P) or {} ? (□◊P → ¬□◊□P) 139. S4 ⊬ {} ? (¬□¬□¬□¬P → ¬P), i.e.{} ? (◊□¬□¬P → ¬P) or {} ? (¬□◊□¬P → ¬P) or {} ? (¬□¬□◊P → ¬P) or {} ? (◊□◊P → ¬P) 140. S4 ⊬ {} ? (¬□¬□¬□¬P → ¬□P), i.e.{} ? (◊□¬□¬P → ¬□P) or {} ? (¬□◊□¬P → ¬□P) or {} ? (¬□¬□◊P → ¬□P) or {} ? (◊□◊P → ¬□P) 141. S4 ⊬ {} ? (¬□¬□¬□¬P → ¬¬□¬P), i.e.{} ? (◊□¬□¬P → ¬◊P) or {} ? (¬□◊□¬P → ¬◊P) or {} ? (¬□¬□◊P → ¬◊P) or {} ? (◊□◊P → ¬◊P) 142. S4 ⊬ {} ? (¬□¬□¬□¬P → ¬¬□¬□P), i.e.{} ? (◊□¬□¬P → ¬◊□P) or {} ? (¬□◊□¬P → ¬◊□P) or {} ? (¬□¬□◊P → ¬◊□P) or {} ? (◊□◊P → ¬◊□P) 143. S4 ⊬ {} ? (¬□¬□¬□¬P → ¬□¬□¬P), i.e.{} ? (◊□¬□¬P → ◊□¬P) or {} ? (¬□◊□¬P → ◊□¬P) or {} ? (¬□¬□◊P → ◊□¬P) or {} ? (◊□◊P → ◊□¬P) or {} ? (◊□¬□¬P → ¬□◊P) or {} ? (¬□◊□¬P → ¬□◊P) or {} ? (¬□¬□◊P → ¬□◊P) or {} ? (◊□◊P → ¬□◊P) 144. S4 ⊬ {} ? (¬□¬□¬□¬P → ¬¬□¬□¬□¬P), i.e.{} ? (◊□¬□¬P → ¬◊□¬□¬P) or {} ? (¬□◊□¬P → ¬◊□¬□¬P) or {} ? (¬□¬□◊P → ¬◊□¬□¬P) or {} ? (◊□◊P → ¬◊□¬□¬P) or {} ? (◊□¬□¬P → ¬¬□◊□¬P) or {} ? (¬□◊□¬P → ¬¬□◊□¬P) or {} ? (¬□¬□◊P → ¬¬□◊□¬P) or {} ? (◊□◊P → ¬¬□◊□¬P) or {} ? (◊□¬□¬P → ¬¬□¬□◊P) or {} ? (¬□◊□¬P → ¬¬□¬□◊P) or {} ? (¬□¬□◊P → ¬¬□¬□◊P) or {} ? (◊□◊P → ¬¬□¬□◊P) or {} ? (◊□¬□¬P → ¬◊□◊P) or {} ? (¬□◊□¬P → ¬◊□◊P) or {} ? (¬□¬□◊P → ¬◊□◊P) or {} ? (◊□◊P → ¬◊□◊P) 145. S4 ⊬ {} ? (¬□¬□¬□¬P → ¬□¬□¬□P), i.e.{} ? (◊□¬□¬P → ◊□¬□P) or {} ? (¬□◊□¬P → ◊□¬□P) or {} ? (¬□¬□◊P → ◊□¬□P) or {} ? (◊□◊P → ◊□¬□P) or {} ? (◊□¬□¬P → ¬□◊□P) or {} ? (¬□◊□¬P → ¬□◊□P) or {} ? (¬□¬□◊P → ¬□◊□P) or {} ? (◊□◊P → ¬□◊□P) 93 None of the following modalities is provable. 146. S4 ⊬ {} ? P, 147. S4 ⊬ {} ? □P, 148. S4 ⊬ {} ? (¬□¬P), i.e.{} ? (◊P) 149. S4 ⊬ {} ? (¬□¬□P), i.e.{} ? (◊□P) 150. S4 ⊬ {} ? (□¬□¬P), i.e.{} ? (□◊P) 151. S4 ⊬ {} ? (¬□¬□¬□¬P), i.e.{} ? (◊□¬□¬P) or {} ? (¬□◊□¬P) or {} ? (¬□¬□◊P) or {} ? (◊□◊P) 152. S4 ⊬ {} ? (□¬□¬□P), i.e.{} ? (□◊□P) 153. S4 ⊬ {} ? (¬P), 154. S4 ⊬ {} ? (¬□P), 155. S4 ⊬ {} ? (¬¬□¬P), i.e.{} ? (¬◊P) 156. S4 ⊬ {} ? (¬¬□¬□P), i.e.{} ? (¬◊□P) 157. S4 ⊬ {} ? (¬□¬□¬P), i.e.{} ? (◊□¬P) or {} ? (¬□◊P) 158. S4 ⊬ {} ? (¬¬□¬□¬□¬P), i.e.{} ? (¬◊□¬□¬P) or {} ? (¬¬□◊□¬P) or {} ? (¬¬□¬□◊P) or {} ? (¬◊□◊P) 159. S4 ⊬ {} ? (¬□¬□¬□P), i.e.{} ? (◊□¬□P) or {} ? (¬□◊□P) Properties not required by an S4 model (symmetry : S4 □◊P)) and S4 160. S4 ⊬ {} ? (¬□¬P → □¬□¬P) ({} ? (¬□¬P → □◊P)) , and {} ? (P → □¬□¬P) ({} ? (P → ⊬ {} ? (¬□¬P → ¬□¬□¬P) ({} ? (◊P → ◊□¬P) or {} ? (◊P → ¬□◊P)) are shown above) ⊬ {} ? (□(□P→P) → □P), i.e. Löbs rule (converse well founded) Linear ordering of the worlds is not enough to model S4 161. S4 ⊬ {¬□¬P, ¬□¬¬P} ? Q, i.e. {◊P, ◊¬P} ? Q 94 162. S4 ⊬ {} ? (□(□(P→□P)→P)→P), i.e, .reflexivity+transiticity+converse-wellfounded 163. S4 ⊬ {} ? (□(□A→B) v □(□B→A)) Miscellaneous 164. S4 ⊬ {(R→S), R} ? □S, 165. S4 ⊬ {(R→S), □R} ? □S, 166. S4 ⊬ {□(R→S), R}, ? □S, 167. S4 ⊬ {□R, S} ? □(R→S), 168. S4 ⊬ {□R, S} ? □(R&S), 169. S4 ⊬ {R, □S} ? □(R&S), 170. S4 ⊬ {R, S} ? □(RvS), 171. S4 ⊬ {□(BvC)} ? (□Bv□C), 172. S4 ⊬ {Q→□Y} ? □(X&□(P&Q) → □Y), 173. S4 ⊬ {□((R↔S)&Q), □(RvZ), ¬Z} ? □(SvP) 95 2. Examples in S5: All of the statements provable in S4 are provable in S5 too. AProS proves all the statements listed above (as provable in S4) in S5. Theorems: The examples are organized as follows23: I. Axioms of S5 II. Reduction rules (generated from distinct modalities) III. Miscellaneous examples with other connectives I. Axioms of S5. The axioms of necessitation, distributivity, reflexivity and transitivity are provable in S5, and the proofs are exactly the same as they were for S4. S5 has an additional axiom that we prove here. S5 ⊦□ → □□ S5 ⊦□ □□ Y □ □ □I □ □□ Lob’s axiom is unprovable as AproS shows. II.Properties of S5 models S5 models have accessibility relation that is an equivalence. All of the properties are direct axioms. Other properties such as well-foundedness form statements that are unprovable in S5 (and AProS shows this). III.Distinct modalities of S5 The set of distinct modalities for S5 are: DS ={*P, P | * is one of □, ◊, ¬□, ¬◊ ¬} 23 Only the statements that are provable only in S5 are numbered. 96 All the other modalities reduce to one of these using the following reduction rules. Reduction rules (proved in the next pages) S5 ⊦{} ? □P ↔ □□P, (shown for S4) S5 ⊦{} ? ¬□¬P ↔ ¬□¬¬□¬P (shown for S4) 2. S5 ⊦{} ? □P ↔ ¬□¬□P i.e.{} ? □P↔ ◊□P 3. S5 ⊦{} ? ¬□¬P ↔ □¬□¬P i.e.{} ? ◊P ↔ □◊P Since the rules listed here are bi-implications, the negated versions of these rules are already proves. However, APros was made to prove them since the logics differ in the way their box-rules handle negations which is verified here. The proofs of these are not listed. S5 ⊦{} ? ¬□P↔¬□□P, S5 ⊦{} ? ¬¬□¬P↔¬¬□¬¬□¬P, i.e.{} ? ¬◊P↔¬◊◊P 4. S5 ⊦{} ? ¬□P ↔ ¬¬□¬□P i.e.{} ? ¬□P↔ ¬◊□P 5. S5 ⊦{} ? ¬¬□¬P ↔ ¬□¬□¬P i.e.{} ? ¬◊P ↔ ¬□◊P Modal logics are usually presented using connectives □ and ◊. Because of this, the formulae ¬□P and ◊¬P are syntactically different, and reduction rules list them. These are a subset of those for S4. S5 ⊦{} ? (□¬P↔¬¬□¬P), i.e. {} ? (□¬P↔¬◊P) S5 ⊦{} ? (¬□¬¬P↔¬□P), i.e.{} ? (◊¬P↔¬□P) S5 ⊦{} ? (□¬□¬¬P↔¬¬□¬□P), i.e.{} ? (□◊¬P↔¬◊□P) S5 ⊦{} ? (¬□¬□¬P↔¬□¬□¬P), i.e.{} ? (¬□◊P↔¬□◊P) or {} ? (¬□◊P↔◊□¬P) or {} ? (◊□¬P↔¬□◊P) or {} ? (◊□¬P↔◊□¬P) 97 2. S5 ⊦{□P } ? (¬□¬□P) Y {□P; ¬□P }? ¬□P Y □E {□P, □¬□P } ? ¬□P {□P, ¬□P }? □P ⊥I {□P, □¬□P } ? ⊥ ¬I {□P} ? ¬□¬□P 3. S5 ⊦{¬□¬P } ? (□¬□¬P) Y {¬□¬P} ? ¬□¬P □I {¬□¬P} ? □¬□¬P The following is a list of application of reduction rules. The list contains one statement for each of the distinct modalities. The proofs are very long and are not listed here. 6. S5 ⊦ {} ? □P↔□□□P 7. S5 ⊦ {} ? □P↔□□□□P 8. S5 ⊦ {} ? ¬□¬P↔¬□¬¬□¬¬□¬P 9. S5 ⊦ {} ? ¬□¬P↔¬□□□¬P 10. S5 ⊦ {} ? □□¬□¬□¬□¬¬□¬□□□□□□P↔□□□□¬□¬¬□¬¬□¬¬□¬□¬□¬□P 11. S5 ⊦ {} ? □□□□□□¬□¬□¬□¬¬□¬□□¬□¬P↔□□□□¬□¬¬□¬¬□¬¬□¬□¬□¬□¬□¬P 12. S5 ⊦ {} ? ¬□□□□□□¬□¬□¬□¬¬□¬□¬¬□¬P↔¬□□□□¬□¬¬□¬¬□¬¬□¬□¬□¬□¬P 13. S5 ⊦ {} ? ¬□□¬□¬□¬□¬¬□P↔¬□□□□¬□¬¬□¬¬□¬¬□¬□¬□P 14. S5 ⊦ {} ? ¬□P↔¬□□□P 15. S5 ⊦ {} ? ¬¬□¬P↔¬¬□¬¬□¬¬□¬P 16. S5 ⊦ {} ? ¬□□¬□¬□¬□¬¬□¬□□P↔¬□□□□¬□¬¬□¬¬□¬¬□¬□¬□¬□P 98 17. S5 ⊦ {} ? ¬□□¬□¬□¬□¬¬□¬□□¬□¬P↔¬□□□□¬□¬¬□¬¬□¬¬□¬□¬□¬□¬□¬P 18. S5 ⊦ {} ? ¬¬□□¬□¬□¬□¬¬□¬□¬¬□¬P↔¬¬□□□□¬□¬¬□¬¬□¬¬□¬□¬□¬□¬P 19. S5 ⊦ {} ? ¬¬□□¬□¬□¬□¬¬□P↔¬¬□□□□¬□¬¬□¬¬□¬¬□¬□¬□P Some distinct modalities are not equivalent, but one may imply the other The proofs of some of these are given in the next pages, namely, 20, 21, 22 and 30. 20. S5 ⊦ {} ? P → □¬□¬P 21. S5 ⊦ {} ? ¬□¬□P → ¬□¬P S5 ⊦ {} ? □P → P (shown for S4) S5 ⊦ {} ? P → ¬□¬P (shown for S4) S5 ⊦ {} ? □P → ¬□¬P (shown for S4) 22. S5 ⊦ {} ? ¬□¬□P → □¬□¬P 23. S5 ⊦ {} ? ¬□¬□P → P S5 ⊦ {} ? □P → □¬□¬P (shown for S4) 24. S5 ⊦ {} ? ¬□¬□¬P → ¬P 25. S5 ⊦ {} ? ¬□¬□¬P → ¬¬□¬□P S5 ⊦ {} ? ¬P → ¬□P S5 ⊦ {} ? ¬¬□¬P → ¬P S5 ⊦ {} ? ¬¬□¬P → ¬□P 26. S5 ⊦ {} ? ¬¬□¬P → ¬¬□¬□P 27. S5 ⊦ {} ? ¬P → ¬¬□¬□P 28. S5 ⊦ {} ? ¬□¬□¬P → ¬□P 29. S5 ⊦ {} ? ¬□¬□P → ¬□¬P 30. S5 ⊦ {} ? ¬□¬□P → P 99 20. S5 ⊦ {} ? P → □¬□¬P Y {□P, ¬□¬□¬P, ¬□¬P } ? ¬□¬P Y As before {□P, ¬□¬□¬P, ¬□¬P } ? ¬□¬□¬P P } ? □¬P {□P, ¬□¬□¬P, ¬□¬P } ? □¬□¬P P } ? □¬P P } ? □¬P □I \ {□P, ¬□¬□¬P, ¬□¬P } ? ⊥ \ P } ?{□ □¬ P,P¬□¬□¬ P} ? □¬P {□P, ¬□¬□¬P } ? ¬□¬P ⊥I ¬I \ ⊥I \ P } ? □¬P {□P, ¬□¬□¬P } ? ⊥ \¬I {P} ? □¬□¬P 21. S5 ⊦ {} ? ¬□¬□P → ¬□¬P Y Y {¬□¬□P,□¬P, P } ? P {¬□¬□P,¬P, □ P } ? ¬P □I □I {¬□¬□P,□¬P, □ P } ? P P } ? □¬P \ P } ? □¬P {¬□¬□P,□¬P, □ P } ? ¬P {¬□¬□P,□¬P, □ P } ? ⊥ P } ? □¬P ¬I \ \ P } ? □¬P {¬□¬□P,□¬P } ? ¬□ P Y P } ? □¬P {¬□¬□P,□¬P } ? □¬□ P {¬□¬□P,□¬P } ? ¬□¬□ P P } \? P □¬} ?P □¬P {¬□¬□P,□¬P } ? ⊥ \ \ ¬I {¬□¬□P} ? ¬□¬P 22. S5 ⊦ {} ? ¬□¬□P → □¬□¬P The proof of this is simply □I applied to the proof above. 100 □I \ ⊥I ⊥I Y 30. S5 ⊦ {} ? ¬□¬□P → P { ¬□¬□P, ¬P; P } ? P Y As before P } ? □¬P {□P, ¬□¬□P, ¬P } ? ¬P {□P, ¬□¬□P, ¬P } ? P P } ? □¬P P } ? □¬P \ \ {□P, ¬□¬□P, ¬P } ? ⊥ P } ?{¬□¬□ □¬P P, ¬P } ? ¬□P {¬□¬□P, ¬P } ? □P □E \ } ? □¬P ¬I \ ⊥I {¬□¬□P, ¬P } ? ⊥ \¬I {¬□¬□P} ? P 31. S5 ⊦ {}? (□(□A → B) v □(□B → A)) The proof is not included, but it can be seen that this follows from distributivity and □-elimination. 101 ⊥I Unprovable Statements: All of the statements stated below are unprovable in S4 as well (as AProS shows). Distinct modalities are unboxed, □, ◊ and their negations P is distinct from other modalities 1. S5 ⊬ 2. S5 ⊬ {} ? (P ↔ ¬□¬P), 3. S5 ⊬ {} ? (P ↔ ¬P), 4. S5 ⊬ {} ? (P ↔ ¬¬□¬P), i.e.{} ? (P ↔ ¬◊P) 5. S5 ⊬ {} ? (P ↔ ¬□P), {} ? (P ↔ □P), i.e.{} ? (P ↔ ◊P), □P is distinct from other modalities 6. S5 ⊬ {} ? (□P ↔ ¬□¬P), i.e.{} ? (□P ↔ ◊P) 7. S5 ⊬ {} ? (□P ↔ ¬P), 8. S5 ⊬ {} ? (□P ↔ ¬□P) , 9. S5 ⊬ {} ? (□P ↔ ¬¬□¬P), i.e.{} ? (□P ↔ ¬◊P) ◊P is distinct from other modalities 10. S5 ⊬ {} ? (¬□¬P ↔ ¬¬□¬P), i.e.{} ? (◊P ↔ ¬◊P) i.e. ¬P is distinct from other modalities 11. S5 ⊬ 12. S5 ⊬ {} ? (¬P ↔ ¬□P), {} ? (¬P ↔ ¬¬□¬P), i.e.{} ? (¬P ↔ ¬◊P) ¬□P is distinct from other modalities 13. S5 ⊬ {} ? (¬□P ↔ ¬¬□¬P), i.e.{} ? (¬□P ↔ ¬◊P) One side inclusions of the distinct modalities 14. S5 ⊬ {} ? (P → □P), 15. S5 ⊬ {} ? (P → ¬P), 102 16. S5 ⊬ 17. S5 ⊬ {} ? (P → ¬□P), {} ? (P → ¬¬□¬P), i.e.{} ? (P → ¬◊P) □P is distinct from other modalities 18. S5 ⊬ {} ? (□P → ¬P), 19. S5 ⊬ {} ? (□P → ¬□P) , 20. S5 ⊬ {} ? (□P → ¬¬□¬P), i.e.{} ? (□P → ¬◊P) ◊P is distinct from other modalities 21. S5 ⊬ {} ? (¬□¬P → ¬¬□¬P), i.e.{} ? (◊P → ¬◊P) ¬P is distinct from other modalities 22. S5 ⊬ {} ? (¬P → ¬¬□¬P), i.e.{} ? (¬P → ¬◊P) ¬□P is distinct from other modalities 23. S5 ⊬ {} ? (¬□P → ¬¬□¬P), i.e.{} ? (¬□P → ¬◊P) None of the modalities are provable outright 24. S5 ⊬ {} ? P 25. S5 ⊬ {} ? (□P), 26. S5 ⊬ {} ? (¬□¬P), i.e.{} ? (◊P) 27. S5 ⊬ {} ? ¬P 28. S5 ⊬ {} ? (¬□P), 29. S5 ⊬ {} ? (¬¬□¬P), i.e.{} ? (◊P) Properties not required by an S5 model Löbs rule (converse well founded) {} ? □( □P→P) → □P 103 Miscellaneous 30. S5 ⊬ {(R→S), R } ? □S 31. S5 ⊬ {(R→S), □R } ? □S, 32. S5 ⊬ {□(R→S), R } ? □S 33. S5 ⊬ {□R, S } ? (R→S) 34. S5 ⊬ {□R, S } ? (R & S) 35 – 170. Apros was also made to prove that the distinct modalities of S4 which still hold for S5, and their one side inclusions. Example: S5 here. ⊬ {} ? (¬□¬□¬□¬P → □P). We do not list these 104 3. Examples in GL: Theorems: The examples are organized as follows: I. Axioms of GL. II. Gödel’s second incompleteness theorem, and variants (□¬□A proves anything). III. There are an infinite number of distinct modalities, so we show a few of the reduction rules. IV. Miscellaneous examples with other connectives. I. Axioms of GL. Necessitation: For any tautology A, GL ⊦ □A (proof is the same as that of S4). Distributivity: □(A→B) proves □A→□B (proof is the same as that of S4). Löb’s axiom: □(□A→A) →□A 1. GL ⊦ {□(□A→A) } ? □A Y Y {(□A A),□A,A } ? □A {(□A A),□A,A } ? A E {□(□A A), (□A A),□A } ? A □I {□(□A A)} ? □A Converse: GL ⊦ {} ? (□A → □(□A→A)) 2. GL ⊦ {□A } ? □(□A→A) Y {□(□A A), □A, A } ? A I {□(□A A), □A, A } ? (□AA) □I {□A } ? □(□A A) 105 II.Gödel’s second incompleteness theorem, and variants 3.{} ? □¬□(A&¬A) → □(A&¬A) 3. GL ⊦ {□¬□(A&¬A)} ? □(A&¬A) Y Y {□¬□(A&¬A), ¬□(A&¬A), □(A&¬A)} ? □(A&¬A) {□¬□(A&¬A), ¬□(A&¬A), □(A&¬A)} ? ¬□(A&¬A) ⊥I {□¬□(A&¬A), ¬□(A&¬A), □(A&¬A), ¬ (A&¬A)} ? ⊥ {□¬□(A&¬A), ¬□(A&¬A), □(A&¬A)} ? A&¬A ¬E □I {□¬□(A&¬A)} ? □(A&¬A) 4. GL ⊦ {□¬□P} ? □P Y Y {□¬□P, ¬□P, □P} ? □P {□¬□P, ¬□P, □P} ? ¬□P ⊥I {□¬□P, ¬□P, □P, ¬P } ? ⊥ ¬E {□¬□P, ¬□P, □P} ? P □I {□¬□P} ? □P 106 5. GL ⊦ {¬□P} ? ¬□¬□P Y Y {□¬□P, ¬□P, □P, ¬P } ? ¬□P {□¬□P, ¬□P, □P, ¬P } ? □P ⊥I {□¬□P, ¬□P, □P, ¬P } ? ⊥ ¬E Y {□¬□P, ¬□P □P } ? P □I {□¬□P, ¬□P } ? ¬□P {□¬□P, ¬□P,} ? □P ⊥I {¬□P, □¬□P} ? ⊥ ¬I {¬□P} ? ¬□¬□P 6. GL ⊦ {¬□¬P} ? ¬□¬□¬P ⊥I ¬I Y □I Y {□¬□¬P, ¬□¬P, □¬P, P } ? ¬□¬P {□¬□¬P, ¬□¬P, □¬P, P } ? □¬P ⊥I {□¬□¬P, ¬□¬P, □¬P, P } ? ⊥ Y {□¬□¬P, ¬□¬P } ? ¬□¬P {□¬□¬P, ¬□¬P □¬P } ? ¬ P {□¬□¬P, ¬¬□P} ? □¬P {¬□¬P, □¬□¬P} ? ⊥ {¬□¬P} ? ¬□¬□¬P 107 ¬I 7. GL ⊦ {□¬□P} ? □¬□Q Y Y {□¬□P, ¬□P, □¬□Q, ¬□Q, □P, ¬P } ? ¬□P {□¬□P, ¬□P, □¬□Q, ¬□Q, □P, ¬P } ? □P ⊥I {□¬□P, ¬□P, □¬□Q, ¬□Q, □P, ¬P } ? ⊥ ¬E {□¬□P, ¬□P, □¬□Q, ¬□Q □P } ? P Y □I {□¬□P, ¬□P, □¬□Q, ¬□Q } ? ¬□P {□¬□P, ¬□P, □¬□Q, ¬□Q } ? □P ⊥I {□¬□P, ¬□P, □¬□Q, ¬□Q } ? ⊥ ¬I {□¬□P, ¬□P, □¬□Q} ? ¬□Q □I {□¬□P} ? □¬□Q 8. GL ⊦ {□¬□P} ? □Q Y Y {□¬□P, ¬□P, ¬Q, □Q, □P, ¬P } ? ¬□P {□¬□P, ¬□P, ¬Q, □Q, □P, ¬P } ? □P ⊥I {□¬□P, ¬□P, ¬Q, □Q, □P, ¬P } ? ⊥ ¬E {□¬□P, ¬□P, ¬Q, □Q □P } ? P Y □I {□¬□P, ¬□P, □Q, ¬Q } ? ¬□P {□¬□P, ¬□P, □Q, ¬Q } ? □P ⊥I {□¬□P, ¬□P, ¬Q, □Q } ? ⊥ ¬E {□¬□P, ¬□P, □Q} ? Q □I {□¬□P} ? □Q 108 9. GL ⊦ {□¬□¬P} ? □¬P Y Y {□¬□¬P, ¬□¬P, □¬P , P} ? □¬P {□¬□¬P, ¬□¬P, □¬P , P} ? ¬□¬P ⊥I {□¬□¬P, ¬□¬P, □¬P , P} ? ⊥ ¬I {□¬□¬P, ¬□¬P, □¬P } ? ¬P □I {□¬□¬P} ? □¬P 10. GL ⊦ {¬□¬P, □Q} ? ¬□¬Q Y Y {□Q, □¬Q , Q, ¬Q, P} ? Q {□Q, □¬Q , Q, ¬Q, P} ? ¬Q ⊥I {□Q, □¬Q , Q, ¬Q, P} ? ⊥ ¬I { □Q, □¬Q , Q, ¬Q } ? ¬P Y □I {¬□¬P, □Q, □¬Q } ? □¬P {¬□¬P, □Q, □¬Q } ? □¬P ⊥I {¬□¬P, □Q, □¬Q} ? ⊥ ¬I {¬□¬P, □Q} ? ¬□¬Q 109 11. GL ⊦ {¬□P, □Q} ? ¬□¬Q ¬I Y Y {{□¬P, ¬P, □Q, □¬Q , Q, ¬Q, P} ? Q {{□¬P, ¬P, □Q, □¬Q , Q, ¬Q, P} ? ¬Q ⊥I {{□¬P, ¬P, □Q, □¬Q , Q, ¬Q, P} ? ⊥ ¬E {□¬P, ¬P, □Q, □¬Q , Q, ¬Q } ? P Y □I {¬□P, □Q, □¬Q } ? ¬□P {□¬P, □Q, □¬Q } ? □P ⊥I {¬□P, □Q, □¬Q} ? ⊥ ¬I {¬□P, □Q} ? ¬□¬Q AProS also proves other variants of this theme such as: 12. GL ⊦ {¬□P, □(A &¬A)} ? Q 13. GL ⊦ {□(A &¬A)} ? □Q 14. GL ⊦ {¬□¬□¬P} ? ¬□¬□¬Q24 15. GL ⊦ {¬□¬□P} ? (¬□¬□¬□¬P) 16. GL ⊦ {□¬□¬□P} ? □¬□¬P 17. GL ⊦ { ¬□¬□¬□P} ? (¬□¬□¬□¬P) AProS also proves modifications of distributivity and transitivity, and the definition of box such as: These are similar to the proofs in S4. GL ⊦ {R, □S} ? □(RS) GL ⊦ {□R, □S} ? □(RvS) GL ⊦ {□(B&C)} ? □B&□C 24 I am not using diamonds here, since there is no standard format which uses it. 110 GL ⊦ {¬□¬¬□¬P} ? (¬□¬P) GL ⊦ {□(□PP)} ? (□□P□P) GL ⊦ {¬□¬P} ? (¬□¬□¬P) GL ⊦ {□(P & ¬P)}, (□Q) GL ⊦ {□P,¬□¬¬P} ? Q GL ⊦ {□¬P, ¬□¬¬¬P} ? Q GL ⊦ {□P} ? □P GL ⊦ {□R & □S} ? □R, GL ⊦ {□(R & S)} ? □R GL ⊦ {□(R & S)} ? □S GL ⊦ {(□R) v (□S)} ? □(R v S) GL ⊦ {□□B, □((A v ¬A) □A (□BA)) } ? □A GL ⊦ {¬□A, A¬A} ? (¬□¬□A) GL ⊦{¬□(□(A&¬A) (A&¬A))} ? (¬□¬□(A&¬A)) GL ⊦{¬□(A&¬A)} ? (¬□(□(A&¬A) (A&¬A))) GL ⊦ {¬□(A&¬A)} ? (¬□¬□(A&¬A)) GL ⊦ {□¬□(A&¬A)} ? □R GL ⊦ {□¬□(A&¬A)} ? □¬□¬R GL ⊦ {□A} ? □(□AA) GL ⊦ {□(A&¬A)} ? □(¬A) GL ⊦ {□(A&¬A)} ? □¬□¬A GL ⊦ {(A&¬A)} ? (¬□¬P) GL ⊦ {□R, □S} ? □(RS) 111 GL ⊦ {R, □S} ? □(RS) GL ⊦ {□R, □S} ? □□(RS) GL ⊦ {□R, □S} ? □(R&S) GL ⊦ {□R, □S} ? □(RvS) GL ⊦ {□R, S } ? □(RvS) GL ⊦ {R, □S} ? □(RvS) There are infinite distinct modalities; the following lists a few. GL ⊦ {□P} ? □□□P GL ⊦ {¬□¬□P} ? ¬□¬□□□P GL ⊦ {□¬□¬□P} ? □¬□¬□□□P GL ⊦ {¬□□□¬□¬P} ? ¬□¬□¬Q GL ⊦ {□¬□¬□P} ? □¬□¬Q GL ⊦ {¬□¬□¬□P} ? ¬□¬□¬Q GL ⊦ {□¬□¬□¬P} ? □¬□¬Q GL ⊦ {¬□¬□¬□¬P} ? ¬□¬□¬Q GL ⊦ {□¬□¬□¬□P} ? □¬□¬Q GL ⊦ {¬□¬□¬□¬□P} ? ¬□¬□¬Q GL ⊦ {□¬□¬□¬P} ? □¬□¬Q GL ⊦ {¬□¬□¬□¬P} ? ¬□¬□¬Q GL ⊦ {□¬□¬□□□□¬□¬□□¬□□□P} ? □□□¬□¬Q 112 Unprovable statements: The following list is based on properties of models. Reflexivity 1. GL ⊬ 2. GL ⊬ {P} ? ¬□¬P i.e. {P} ? ◊P 3. GL ⊬ {□P, ¬P} ? Q 4. GL ⊬ {□P} ? ¬□¬P i.e, {□P} ? ◊P 5. GL ⊬ {} ? □(□PP) 6. GL ⊬ {} ? P□(□PP) 7. GL ⊬ {} ? □P□¬□¬P 8. GL ⊬ {} ? □P□¬□¬□P 9. GL ⊬ {} ? ¬□¬□P ¬□¬P {□P} ? P 10. GL ⊬ {} ? □P ¬□¬□¬□¬P 11. GL ⊬ {} ? ¬□¬□¬□¬P ¬□¬P 12. GL ⊬ {} ? □¬□¬□P ¬□¬P Symmetry 13. GL ⊬ {¬□¬P} ? □¬□¬P i.e. {◊P} ? □◊P Multiple properties 14. GL ⊬ {} ? □(□(P□P) P) P 15. GL ⊬ {} ? □(□AB) v □(□BA) 16. GL ⊬ {¬□¬P, ¬□¬¬P} ? Q 113 Distinct modalities P, ¬P, □P, □¬P, ¬□P, ¬□¬P, □¬□P … are all distinct (this is just a small subset) 17. GL ⊬ 18. GL ⊬ {} ? P↔□P 19. GL ⊬ {} ? P↔□¬P 20. GL ⊬ {} ? P↔¬□P 21. GL ⊬ {} ? P↔¬□¬P 22. GL ⊬ {} ? P↔□¬□P 23. GL ⊬ {} ? ¬P↔□P 24. GL ⊬ {} ? ¬P↔□¬P 25. GL ⊬ {} ? ¬P↔¬□P 26. GL ⊬ {} ? ¬P↔¬□¬P 27. GL ⊬ {} ? ¬P↔□¬□P 28. GL ⊬ {} ? □P↔□¬P 29. GL ⊬ {} ? □P↔¬□P 30. GL ⊬ {} ? □P↔¬□¬P 31. GL ⊬ {} ? □P↔□¬□P 32. GL ⊬ {} ? ¬□P↔¬□¬P 33. GL ⊬ {} ? ¬□P↔□¬□P 34. GL ⊬ {} ? ¬□¬P↔□¬□P {} ? P↔¬P All the above are totally distinct 114 35. GL ⊬ {} ? P¬P 36. GL ⊬ {} ? P□P 37. GL ⊬ {} ? P□¬P 38. GL ⊬ {} ? P¬□P 39. GL ⊬ {} ? P¬□¬P 40. GL ⊬ 41. GL ⊬ {} ? ¬P□P 42. GL ⊬ {} ? ¬P□¬P 43. GL ⊬ {} ? ¬P¬□P 44. GL ⊬ {} ? ¬P¬□¬P 45. GL ⊬ {} ? ¬P□¬□P 46. GL ⊬ {} ? □P□¬P 47. GL ⊬ {} ? □P¬□P 48. GL ⊬ {} ? □P¬□¬P 49. GL ⊬ {} ? □P□¬□P 50. GL ⊬ {} ? ¬□P¬□¬P 51. GL ⊬ {} ? ¬□P□¬□P 52. GL ⊬ {} ? ¬□¬P□¬□P {} ? P□¬□P These modalities are not provable outright. 53. GL ⊬ {} ? P 115 54. GL ⊬ {} ? □P 55. GL ⊬ {} ? ¬□¬P 56. GL ⊬ {} ? ¬P 57. GL ⊬ {} ? ¬□P 58. GL ⊬ {} ? □¬P 59. GL ⊬ {□(A&¬A) } ? ¬□¬□(A&¬A) 60. GL ⊬ {(RS),R} ? □S 61. GL ⊬ {(RS), □R} ? □S 62. GL ⊬ {□(RS),R} ? □S 63. GL ⊬ {□R, S} ? □(RS) 64. GL ⊬ {□A□B} ? □(AB) 65. GL ⊬ {□R, S} ? □(R&S) 66. GL ⊬ {R, □S} ? □(R&S) 67. GL ⊬ {R, S} ? □(RvS) 116 References: [1] AProS http://www.phil.cmu.edu/projects/apros/ [2] Boolos, G., “The Logic of Provability”, Cambridge University Press, Cambridge, 1993. [3] Byrnes, J., “Proof Search and Normal Forms in Natural Deduction”, PhD Thesis. Carnegie Mellon University (1999). [4] Gödel, K., “Über Formal Unentscheidbare Sätze der Principia Mathematica und Verwandter Systeme I,” Monatshefte für Mathematik und Physik, Vol. 38 (1931): 173-198 [5] Hilbert, D. and Bernays, P. “Grundlagen der Mathematik”, Vol 2. Berlin, Heidelberg, New York, Springer-Verlag (1939). [6] Löb, M.H., “Solution of a Problem of Leon Henkin,” Journal of Symbolic Logic, Vol. 20 (1955): 115-118 [7] Prawitz, D., “Natural Deduction. A Proof-Theoretic Study”, 1965, ABC i Symbolisk Logik, 1975, 2:a uppl (1991): 74-80 [8] Solovay, R.M., “Provability Interpretations of Modal Logic,” Israel Journal of Mathematics, Vol. 25 (1976): 287-304. [9] Sieg, W., Cittadini, S., “Normal Natural Deduction Proofs (in Non-classical logics),” Mechanizing Mathematical Reasoning, LNAI 2605, 169-191, 2005. [10] Sieg, W., Byrnes, J. “Normal Natural Deduction Proofs (in classical logic),” Studia Logica 60, 67-106, 1998. [11] Troelstra, A.S., Schwichtenberg, H., “Basic Proof Theory.” In series Cambridge Tracts in Theoretical Computer Science, Cambridge University Press, (1996). 117