A Simplification of the Completeness Proofs for Guaspari and Solovay's R

Alternative proofs for Guaspari and Solovay's completeness theorems for R are presented. R is an extension of the provability logic L and was developed in order to study the formal properties of the provability predicate of PA occurring in sentences that may contain connectives for witness comparison. (The primary example of sentences involving witness comparison is the Rosser sentence.) In this article the proof of the Kripke model completeness theorems employs tail models, as introduced by Visser, instead of the more usual finite Kripke models. The use of tail models makes it possible to derive arithmetical completeness from Kripke model completeness by literally embedding Kripke models into PA. Our arithmetical completeness theorem differs slightly from the one proved by Guaspari and Solovay, and it also forms a solution to the problem (advanced by Smoryήski) of obtaining a completeness result with respect to a variety of orderings. Introduction This paper deals with the completeness proofs given in [2] for the theory R, which is an extension of the provability logic L introduced in [5]. R is formulated in a language containing < and < as connectives for witness comparison, which enables one to study the provability predicate as it occurs in formulas like Pr(A) < P r ( r £ n ) , defined as 3 x ( P r o o f ( x , Γ ^ π ) Λ Vy < x -iProoff.y,'"/?')). Rosser sentences are good examples of this kind of formula. In Section 1 we describe the theories L and R and the alternative complete*The treatment of the Kripke model completeness theorem given in Section 2 has benefited much from [3]. Albert Visser's suggestions, corrections and remarks also made a substantial contribution to this paper. Received June 30, 1986; revised December 8, 1988 COMPLETENESS PROOFS 45 ness proofs for L given by Visser. Our main object is to give similar alternative proofs for R. Sections 2 and 3 deal with the Kripke model, and arithmetical, completeness proofs, respectively. Only elementary facts from modal logic and PA are used. Formally, no knowledge of provability logic is presupposed. For background knowledge, see [1] or the more up-to-date and up-tempo [4]. For that matter, the original articles referred to above are highly readable themselves. Some of the details omitted in this paper can be found in [10]. 1 The theories L and R For convenience we summarize Solovay's completeness theorems from [5] and give definitions of some relevant notions. 1.1 Definition L is the following theory in <£, the language of propositional modal logic (A,B,... denote well-formed formulas of <£): Axioms: Boolean tautologies Ώ(A-+B)-+ (ΠA-+ ΏB) ΠA-> UUA Π{ΠA^>A)-* UA Rules: modus ponens A/ΠA. (Some other names for this system found in the literature are PrL, G, and GL.) 1.2 Theorem (Kripke model completeness) For any sentence A of <£: L h A iff A is valid in all finite, transitive, irreflexive (tree-ordered) Kripke models iff A holds at the root of all finite, transitive, irreflexive (tree-ordered) Kripke models. 1.3 Definition An arithmetical interpretation * of £ is a map from formulas of £ to sentences of PA satisfying: for all atomic p: p* is an arithmetical sentence ± = 0 = 1; T* = 0 = 0 ( ) * commutes with Boolean connectives (UA)* = Pr(A*~). 1.4 Theorem (Arithmetical completeness) For any sentence A of £: L V A iff for all* (PA h A*). In other words, L axiomatizes the schemata provable about the provability predicate Pr(x) in PA. We also have a system L which axiomatizes the true schemata about Pr(x): 1.5 Definition L is the following system of modal logic: Axioms: All theorems of L UA^A Rules: modus ponens. 1.6 Theorem For any sentence A of £ : L h A iff for all* A* is true.