Recursive model theory is supposed to be the study of the effectiveness of constructions and theorems in model theory. This often involves getting “effective” versions of various classical model-theoretic notions. The traditional way of doing this is to restrict attention to recursive models, and recursive isomorphisms between them, etc. Thus for example the following definition appears in the literature (in [3] and [1]).

Definition. Given a recursive model A and an n Є ω, a subset R ⊆ An is called intrinsically r.e. provided that for every recursive model B ≈ A, the isomorphic image in B of R is an r.e. subset of Bn.

It is clear that if R is definable by a (recursive, infinitary) Σ10 formula (with finitely many parameters from A), then R is intrinsically r.e. It seems natural for the converse to be true. Indeed, provided that (A, R) is sufficiently “regular” in a sense made precise in a theorem of Ash and Nerode (see [3]), the converse is true. However, if we drop the (rather strong) regularity conditions, there exist “pathological” examples of intrinsically r.e. relations which are not definable by a Σ10 formula (see [7]).

In this paper, we suggest a rather different approach to studying the effectiveness of model theory, an approach we have dubbed “effective model theory”. The basic idea is to allow arbitrary nonrecursive models, but to require all notions to be relativized to the complexity of the models involved. (Much the same notion has been used in [2] under the name “relatively recursive model theory”.) Thus for example we have the following effective model theory version of the property of being intrinsically r.e.

This paper establishes another very general completeness result for the logics within the field of K4. With each finite transitive frame ℭ we may associate a formula — Bℭ which validates just those frames ℑ in which ℭ is not in a certain sense embeddable (to be exact, ℭ is not the p-morphic image of any subframe of ℑ. By a subframe logic we mean the result of adding such formulas as axioms to K4. The general result is that each subframe logic has the finite model property.

There are a continuum of subframe logics and they include many of the standard ones, such as T, S4, S4.3, S5 and G. It turns out that the subframe logics are exactly those complete for a condition that is closed under subframes (any subframe of a frame satisfying the condition also satisfies the condition). As a consequence, every logic complete for a condition closed under subframes has the finite model property.

It is ascertained which of the subframe logics are compact. It turns out that the compact logics are just those whose axioms express an elementary condition. Tests are given for determining whether a given axiom expresses an elementary condition and for determining what it is in case it does.

In one respect the present general completeness result differs from most of the others in the literature. The others have usually either been what one might call logic based or formula based. They have usually either been to the effect that all of the logics containing a given logic are complete or to the effect that all logics whose axioms come from a given syntactically characterized class of formulas are complete. The present result is, by contrast, what one might call frame based. The axioms of the logics to be proved complete are characterized most directly in terms of their connection with certain frames.

This paper contains a rigorous treatment of the species of implication described in [8] and [9], where it was given the name of connexive implication. A brief historical survey will lay bare its roots in antiquity, and it will be shown that none of the well-known systems of propositional logic serves to formalize it.1 In this paper a new system of ‘connexive’ logic will be presented, the system being shown to be (a) consistent, (b) independent of two-valued logic, (c) Post-complete.

The principal aim of the present paper is to sketch some mathematical applications of the work of the Hilbert school in the foundations of mathematics. Most of them depend on the ε-theorems of Hilbert-Bernays II [6], or of Ackermann [1], but instead the work of Gentzen [5] or Schütte [25], which does not use the ε-symbol, could have been applied.

The avowed purpose of all this work is described in the introduction to volume I of Hilbert-Bernays: the consistency of the usual principles of proof is to be established by means of finitist methods, or at least, by means of methods which are more “evident” or more “constructive” than the principles under discussion. This formulation seems to have several defects:

• (1) Since the notion of constructive proof is vague, the whole formulation of the program is vague; and though an exact formulation constitutes, of course, an interesting problem for the logician, because of this vagueness the mathematician does not find the program attractive.

• (2) The formulation does not cover too well the actual substance of the material contained in [6]; e.g. the ε-theorems for the predicate calculus go far beyond establishing mere consistency of the predicate calculus. It is significant that it is precisely these theorems which lead to the more interesting applications, e.g. the use of the ε-theorems with equality in the solution of Hilbert's 17th problem given below.

The concept of confirmation occupies a central position in the methodology of empirical science. For it is the distinctive characteristic of an empirical hypothesis to be amenable, at least in principle, to a test based on suitable observations or experiments; the empirical data obtained in a test—or, as we shall prefer to say, the observation sentences describing those data—may then either confirm or disconfirm the given hypothesis, or they may be neutral with respect to it. To say that certain observation sentences confirm or disconfirm a hypothesis, does not, of course, generally mean that those observation sentences suffice strictly to prove or to refute the hypothesis in question, but rather that they constitute favorable, or unfavorable, evidence for it; and the term “neutral” is to indicate that the observation sentences are either entirely irrelevant to the hypothesis, or at least insufficient to strengthen or weaken it.

We characterize the collapse of Buss' bounded arithmetic in terms of the provable collapse of the polynomial time hierarchy. We include also some general model-theoretical investigations on fragments of bounded arithmetic.

In this article we shall construct intuitionistic analogues to the main systems of classical tense logic. Since each classical modal logic can be gotten from some tense logic by one of the definitions

• (i) □ p ≡ p ∧ Gp ∧ Hp, ◇p ≡ p ∨ Fp ∨ Pp; or,

• (ii) □ p ≡ p ∧ Gp, ◇p = p ∨ Fp

(see [5]), we shall find that our intuitionistic tense logics give us analogues to the classical modal logics as well.

We shall not here discuss the philosophical issues raised by our logics. Readers interested in the intuitionistic view of time and modality should see [2] for a detailed discussion.

In §2 we define the Kripke models for IKt, the intuitionistic analogue to Lemmon's system Kt. We then prove the completeness and decidability of this system (§§3–5). Finally, we extend our results to other sorts of tense logic and to modal logic.

In the language of IKt, we have: sentence-letters p, q, r, etc.; the (intuitionistic) connectives ∧, ∨, →, ¬; and unary operators P (“it was the case”), F (it will be the case”), H (“it has always been the case”) and G (“it will always be the case”). Formulas are defined inductively: all sentence-letters are formulas; if X is a formula, so are ¬X, PX, FX, HX, and GX; if X and Y are formulas, so are X ∧ Y, X ∨ Y, and X → Y. We shall see that, in contrast to classical tense logic, F and P cannot be defined in terms of G and H.

The object of this note is to show that there is no finite characteristic matrix for any one of Lewis and Langford's systems.

Theorem I. There is no finite characteristic matrix for any one of the systems S1–S5.

Proof. Let M be a matrix with less than n elements for which all the provable formulas in S1, or S2, or S3, or S4, or S5, are satisfied. Let Fn represent the formula

where ∑ stands for a ∨-chain, and the pi, are variables in any one of the calculi. Using M, there is always at least one summand in Fn where pi and pk have the same value. Therefore, Fn can always be written in the form (a = a)∨ B, and thus will give, for any B, a “designated” value, since the formula (p = p) ∨ q is provable in any one of the systems S1–S5.

Give to any one of the systems S1–S5, the following matrix, due to Henle:

1. Elements: all possible classes formed from the integers 1, 2, 3, …, n.

2. “Designated” element: the class {1, 2, 3, …,n}.

3. Boole-Schröder algebra on the elements.

4.

If it is difficult to give the exact significance of consistency proofs from a classical point of view, in particular the proofs of Gentzen [2, 6], and Novikoff [14], the motivations of these proofs are quite clear intuitionistically. Their significance is then less to give a mere consistency proof than to present an intuitionistic explanation of the notion of classical truth. Gentzen for instance summarizes his proof as follows [6]: “Thus propositions of actualist mathematics seem to have a certain utility, but no sense. The major part of my consistency proof, however, consists precisely in ascribing a finitist sense to actualist propositions.” From this point of view, the main part of both Gentzen's and Novikoff's arguments can be stated as establishing that modus ponens is valid w.r.t. this interpretation ascribing a “finitist sense” to classical propositions.

In this paper, we reformulate Gentzen's and Novikoff's “finitist sense” of an arithmetic proposition as a winning strategy for a game associated to it. (To see a proof as a winning strategy has been considered by Lorenzen [10] for intuitionistic logic.) In the light of concurrency theory [7], it is tempting to consider a strategy as an interactive program (which represents thus the “finitist sense” of an arithmetic proposition). We shall show that the validity of modus ponens then gets a quite natural formulation, showing that “internal chatters” between two programs end eventually.

We first present Novikoff's notion of regular formulae, that can be seen as an intuitionistic truth definition for classical infinitary propositional calculus. We use this in order to motivate the second part, which presents a game-theoretic interpretation of the notion of regular formulae, and a proof of the admissibility of modus ponens which is based on this interpretation.

We consider some modal languages with a modal operator D whose semantics is based on the relation of inequality. Basic logical properties such as definability, expressive power and completeness are studied. Also, some connections with a number of other recent proposals to extend the standard modal language are pointed at.

We investigate axiomatizations of Kripke's theory of truth based on the Strong Kleene evaluation scheme for treating sentences lacking a truth value. Feferman's axiomatization KF formulated in classical logic is an indirect approach, because it is not sound with respect to Kripke's semantics in the straightforward sense: only the sentences that can be proved to be true in KF are valid in Kripke's partial models. Reinhardt proposed to focus just on the sentences that can be proved to be true in KF and conjectured that the detour through classical logic in KF is dispensable. We refute Reinhardt's Conjecture, and provide a direct axiomatization PKF of Kripke's theory in partial logic. We argue that any natural axiomatization of Kripke's theory in Strong Kleene logic has the same proof-theoretic strength as PKF. namely the strength of the system ramified analysis or a system of Tarskian ramified truth up to ωω. Thus any such axiomatization is much weaker than Feferman's axiomatization KF in classical logic, which is equivalent to the system of ramified analysis up to ε0.

McKinsey and Tarski [3] described Gödel's proof that the number of Brouwerian-algebraic functions is infinite. They gave an example of a sequence of infinitely many distinct Brouwerian-algebraic functions of one argument, which means that there are infinitely many non-equivalent formulas of one variable in the intuitionistic propositional calculus LJ of Gentzen [1]. However they did not completely characterize such formulas. In § 1 of this note, we define a sequence of basic formulas P∞(X), P0(X), P1(X), … and prove the following theorems.

The (modal) μ-calculus ([14]) is a very powerful extension of modal logic with least and greatest fixed point operators. It is of great interest to computer science for expressing properties of processes such as termination (every run is finite) and fairness (on every infinite run, no action is repeated infinitely often to the exclusion of all others).

The power of the μ-calculus is also evident from a more theoretical perspective. The μ-calculus is a fragment of monadic second-order logic (MSO) containing only formulae that are invariant for bisimulation, in the sense that they cannot distinguish between bisimilar states. Janin and Walukiewicz prove the converse: any property which is invariant for bisimulation and MSO-expressible is already expressible in the μ-calculus ([13]). Yet the μ-calculus enjoys many desirable properties which MSO lacks, like a complete sequent-calculus ([29]), an exponential-time decision procedure, and the finite model property ([25]). Switching from MSO to its bisimulation-invariant fragment gives us these desirable properties.

In this paper we take a classical logician's view of the μ-calculus. As far as we are concerned a new logic should not be allowed into the community of logics without at least considering the standard questions that any logic is bothered with. In this paper we perform this rite of passage for the μ-calculus. The questions we will be concerned with are the following.

If X is a set, [Χ] ω will denote the set of countably infinite subsets of X. ω is the set of natural numbers. If S is a subset of [ω] ω , we shall say that S is Ramsey if there is some infinite subset X of ω such that either [Χ] ω ⊆ S or [Χ] ω ∩ S = 0. Dana Scott (unpublished) has asked which sets, in terms of logical complexity, are Ramsey.

The principal theorem of this paper is: Every Σ 1 1 (i.e., analytic) subset of [ω] ω is Ramsey (for the Σ, Π notations, see Addison [1]). This improves a result of Galvin-Prikry [2] to the effect that every Borel set is Ramsey. Our theorem is essentially optimal because, if the axiom of constructibility is true, then Gödel's Σ2 1 Π2 1 well-ordering of the set of reals [3], having the convenient property that the set of ω-sequences of reals enumerating initial segments is also Σ2 1 ∩ Π2 1, rather directly gives a Σ2 1 ∩ Π2 1 set which is not Ramsey. On the other hand, from the assumption that there is a measurable cardinal we shall derive the conclusion that every Σ 2 1 (i.e., PCA) is Ramsey. Also, we shall explore the connection between Martin's axiom and the Ramsey property.

We compare various notions of algorithmic randomness. First we consider relativized randomness. A set is n-random if it is Martin-Löf random relative to ∅(n − 1). We show that a set is 2-random if and only if there is a constant c such that infinitely many initial segments x of the set are c-incompressible: C(x) ≥ ∣x∣ − c. The ‘only if’ direction was obtained independently by Joseph Miller. This characterization can be extended to the case of time-bounded C-complexity.

Next we prove some results on lowness. Among other things, we characterize the 2-random sets as those l-random sets that are low for Chaitin's Ω. Also, 2-random sets form minimal pairs with 2-generic sets. The r.e. low for Ω. sets coincide with the r.e. K-trivial ones.

Finally we show that the notions of Martin-Löf randomness, recursive randomness, and Schnorr randomness can be separated in every high degree while the same notions coincide in every non-high degree. We make some remarks about hyperimmune-free and PA-complete degrees.

There are several conceptions of truth, such as the classical correspondence conception, the coherence conception and the pragmatic conception. The classical correspondence conception, or Aristotelian conception, received a mathematical treatment in the hands of Tarski (cf. Tarski [1935] and [1944]), which was the starting point of a great progress in logic and in mathematics. In effect, Tarski's semantic ideas, especially his semantic characterization of truth, have exerted a major influence on various disciplines, besides logic and mathematics; for instance, linguistics, the philosophy of science, and the theory of knowledge.

The importance of the Tarskian investigations derives, among other things, from the fact that they constitute a mathematical, formal mark to serve as a reference for the philosophical (informal) conceptions of truth. Today the philosopher knows that the classical conception can be developed and that it is free from paradoxes and other difficulties, if certain precautions are taken.

We believe that is not an exaggeration if we assert that Tarski's theory should be considered as one of the greatest accomplishments of logic and mathematics of our time, an accomplishment which is also of extraordinary relevance to philosophy, as we have already remarked.

In this paper we show that the pragmatic conception of truth, at least in one of its possible interpretations, has also a mathematical formulation, similar in spirit to that given by Tarski to the classical correspondence conception.

In this paper we treat the ramified type theory of Russell [6], afterwards adopted by Whitehead and Russell in Principia mathematica [12], so that we may compare Russell's resolution of the semantical antinomies by ramified type theory with the now widely accepted resolution of them by the method of Tarski in [7], [8], [9].

To avoid impredicativity the essential restriction is that quantification over any domain (type) must not be allowed to add new members to the domain, as it is held that adding new members changes the meaning of quantification over the domain in such a way that a vicious circle results. As Whitehead and Russell point out, there is no one particular form of the doctrine of types that is indispensable to accomplishing this restriction, and they have themselves offered two different versions of the ramified hierarchy in the first edition of Principia (see Preface, p. vii). The version in §§58–59 of the writer's [1], which will be followed in this paper, is still slightly different.

To distinguish Russellian types or types in the sense of the ramified hierarchy from types in the sense of the simple theory of types, let us call the former r-types.

There is an r-type i to which the individual variables belong. If β1, β2, …, βm are any given r-types, m ≧ 0, there is an r-type (β1, β2, …, βm)/n to which there belong m-ary functional variables of level n, n ≧ 1. The r-type (α1, α2, …, αm)/k is said to be directly lower than the r-type (β1, β2, …, βm)/n if α1 = β1, α2 = β2, …, αm = βm, k < n.

In this paper we introduce a new hierarchy of sets and operators which we call the REA hierarchy for “recursively enumerable in and above”. The hierarchy is generated by composing (possibly) transfinite sequences of the pseudo-jump operators considered in Jockusch and Shore [1983]. We there studied pseudo-jump operators defined by analogy with the Turing jump as ones taking a set A to A ⊕ for some index e. We would now call these 1-REA operators and will extend them to α-REA operators for recursive ordinals α in analogy with the iterated Turing jump operators (A → A(α) for α < and Kleene's hyperarithmetic hierarchy. The REA sets will then, of course, be the results of applying these operators to the empty set. They will extend and generalize Kleene's H sets but will still be contained in the class of set singletons thus providing us with a new richer subclass of the set singletons which, as we shall see, is related to the work of Harrington [1975] and [1976] on the problems of Friedman [1975] about the arithmetic degrees of such singletons. Their degrees also give a natural class extending the class H of Jockusch and McLaughlin [1969] by closing it off under transfinite iterations as well as the inclusion of [d, d′] for each degree d in the class. The reason for the class being closed under this last operation is that the REA operators include all operators and so give a new hierarchy for them as well as the sets. This hierarchy also turns out to be related to the difference hierarchy of Ershov [1968], [1968a] and [1970]: every α-r.e. set is α-REA but each level of the REA hierarchy after the first extends all the way through the difference hierarchy although never entirely encompassing even the next level of the difference hierarchy.

In [4] Gordon Plotkin considers the problem of characterizing the λ-definable functionals in full type structures. A plausible, but, as we shall see, quite false, conjecture is that a functional is λ-definable ⇔ it is invariant in the sense of Fraenkel and Mostowski. A better guess might be that the set of λ-definable functionals of type σ has a uniform (in σ) definition in type theory over all full type structures (Plotkin explicitly considers a certain sort of definition by means of “logical relations”). More or less generally, one may ask if the following question is decidable.

Given: a functional F in a full type structure over a finite ground domain.

Question: is Fλ-definable?

We call the statement that this question is decidable Plotkin's λ-definability conjecture. We do not know if Plotkin's conjecture is true.

In this note we consider several questions which quickly arise from Plotkin's conjecture. In §1, we ask which type structures satisfy λ-definability = invariance. We construct for each consistent set of equations a model satisfying this equality. From consideration of these models we obtain a number of syntactic corollaries including a reduction of βη-conversion to βη-conversion at a single type (Theorem 3), an “ω-rule” for βη-conversion (Proposition 6) and consistency with βη-conversion (Proposition 8), definability and indefinability results for versions of Kronecker's δ (Propositions 10 and 11), and a decidability result for the problem of inter-definability among combinators.