In this paper we continue, from [2], the development of provably recursive analysis, that is, the study of real numbers defined by programs which can be proven to be correct in some fixed axiom system S. In particular we develop the provable analogue of an effective operator on the set of recursive real numbers, namely, a provably correct operator on the set of provably recursive real numbers. In Theorems 1 and 2 we exhibit a provably correct operator on which is discontinuous at 0; we thus disprove the analogue of the Ceitin-Moschovakis theorem of recursive analysis, which states that every effective operator on is (effectively) continuous. Our final theorems show, however, that no provably correct operator on can be proven (in S) to be discontinuous.

In [1] and [2] D. Lewis formulates his counterfactual logic VC as follows. The language contains the connectives ∧, ∨, ⊃, ¬ and the binary connective ≤. A ≤ B is read as “A is at least as possible as B”. The following connectives are defined in terms of ≤.

A < B: = ¬(B ≤ A) (it is more possible that A than that B).

◊ A ≔ ¬(⊥ ≤ A) (⊥ is the false formula; A is possible).

□ A ≔ ⊥ ≤ ¬A (A is necessary).

(if A were the case, then B would be the case).

(if A were the case, then B might be the case).

and are two counterfactual conditional operators. (AB) iff ¬(A ¬B).

The following axiom system VC is presented by D. Lewis in [1] and [2]: V: (1) Truthfunctional classical propositional calculus.

The problem of changing the cofinality of a measurable cardinal to ω with the help of an iterated ultrapower construction has been introduced in [Bu] and more completely studied in [De]. The aim of this paper is to investigate how the construction above has to be changed to obtain an uncountable cofinality for the (previously) measurable cardinal.

A forcing approach of this question has been developed by Magidor in [Ma]. Just as in the countable case with Prikry forcing, it turns out that the needed hypothesis and the models constructed are the same in both techniques. However the ultrapowers yield a solution which may appear as more effective. In particular the sequence used to change the measurable cardinal into a cardinal of cofinality α has the property that for any β < α the restriction to β of this sequence can be used to change the cofinality of the (same) measurable cardinal to β.

The result we prove is as follows:

Theorem. Assume that α is a limit ordinal, that (Uβ)β<α is a sequence of complete ultrafilters on κ > α in the model N0, andfor B included in α let NB be the ultrapower of N0 by those Uβ which are such that β is in B.

Given two (positive) equivalence relations ~1, ~2 on the set ω of natural numbers, we say that ~1 is m-reducible to ~2 if there exists a total recursive function h such that for every x, y ∈ ω, we have x ~1y iff hx ~2hy. We prove that the equivalence relation induced in ω by a positive precomplete numeration is complete with respect to this reducibility (and, moreover, a “uniformity property” holds). This result allows us to state a classification theorem for positive equivalence relations (Theorem 2). We show that there exist nonisomorphic positive equivalence relations which are complete with respect to the above reducibility; in particular, we discuss the provable equivalence of a strong enough theory: this relation is complete with respect to reducibility but it does not correspond to a precomplete numeration.

From this fact we deduce that an equivalence relation on ω can be strongly represented by a formula (see Definition 8) iff it is positive. At last, we interpret the situation from a topological point of view. Among other things, we generalize a result of Visser by showing that the topological space corresponding to a partition in e.i. sets is irreducible and we prove that the set of equivalence classes of true sentences is dense in the Lindenbaum algebra of the theory.

The permutation method was first applied to NF by Scott; other workers have published results ([2], [3]; or see [4] for a survey). Some of the results proved here have a more metalogical character than most previously yielded by this method. Hinnion and Petry (unpublished, but see [4]) proved that the existence of objects x such that x = {y: x∈y} is consistent with NF. (The significance of this is that the operation that sends x to {y: x∈y} respects ∈ and is thus an embedding.) It is demonstrated below that the existence of such objects is independent of the axioms of NF.

The existence of nontrivial automorphisms of the universe is not an interesting possibility in ZF, since it contradicts wellfoundedness. Similar auguments are not available in NF, however, and it is shown below that if NF + AC for pairs is consistent, then we can consistently add an axiom stating that there is a ∈-automorphism of the universe that is a set of the model. (In the proof given below, the automorphism is in fact of order 2, but natural enrichments of the construction enable one to find automorphisms of other orders with suitable versions of choice as additional hypotheses.)

Ehrenfeucht and Mostowski [3] introduced the notion of indiscernibles and proved that every first order theory has a model with an infinite set of order indiscernibles. Since their work, techniques involving indiscernibles have proved to be extremely useful for constructing models with various specialized properties. In this paper and in a sequel [5], we investigate the effective content of Ehrenfeucht's and Mostowski's result. In this paper we consider the question of which decidable theories have decidable models with infinite recursive sets of indiscernibles. In §1, using some basic facts from stability theory, we show that certain large classes of decidable theories have decidable models with infinite recursive sets of indiscernibles. For example, we show that every ω-stable decidable theory and every stable theory which possesses a certain strong decidability property called ∃Q-decidability have such models. In §2 we construct several examples of decidable theories which have no decidable models with infinite recursive sets of indiscernibles. These examples show that our hypothesis for our positive results in §1 are necessary. Finally in §3 we give two applications of our results. First as an easy application of our results in §1, we show that every ω-stable decidable theory has uncountable models which realize only recursive types. Also our counterexamples in §2 allow us to answer negatively two questions of Baldwin and Kueker [1] concerning the effectiveness of their elimination of Ramsey quantifiers for certain theories.

In [5], we show that in general the problem of finding an infinite set of indiscernibles in a decidable model is recursively equivalent to finding a path through a recursive infinite branching tree. Similarly, we show that the problem of finding an co-type of a set of indiscernibles in a decidable ω-categorical theory is recursively equivalent to finding a path through a highly recursive finitely branching tree.

The number of homogeneous models has been studied in [1] and other papers. But the number of countable homogeneous models of a countable theory T is not determined when dropping the GCH. Morley in [2] proves that if a countable theory T has more than ℵ1 nonisomorphic countable models, then it has such models. He conjectures that if a countable theory T has more than ℵ0 nonisomorphic countable models, then it has such models. In this paper we show that if a countable theory T has more than ℵ0 nonisomorphic countable homogeneous models, then it has such models.

We adopt the conventions in [1]–[3]. Throughout the paper T is a theory and the language of T is denoted by L which is countable.

Lemma 1. If a theory T has more than ℵ0types, then T hasnonisomorphic countable homogeneous models.

Proof. Suppose that T has more than ℵ0 types. From [2, Corollary 2.4] T has types. Let σ be a Ttype with n variables, and T′ = T ⋃ {σ(c1, …, cn)}, where c1, …, cn are new constants. T′ is consistent and has a countable model (, a1, …, an). From [3, Theorem 3.2.8] the reduced model has a countable homogeneous elementary extension . σ is realized in . This shows that every type σ is realized in at least one countable homogeneous model of T. But each countable model can realize at most ℵ0 types. Hence T has at least countable homogeneous models. On the other hand, a countable theory can have at most nonisomorphic countable models. Hence the number of nonisomorphic countable homogeneous models of T is .

In the following, we shall use the languages Lα (α = 0, 1, 2) defined in [2]. We give a brief description of them. For a countable theory T, let K be the class of all models of T. L = L0 is countable.

We use some simple facts about the wtt-degrees of r.e. sets together with a construction to answer some questions concerning the join and meet operators in the r.e. degrees. The construction is that of an r.e. Turing degree a with just one wtt-degree in a such that a is the join of a minimal pair of r.e. degrees. We hope to illustrate the usefulness of studying the stronger reducibility orderings of r.e. sets for providing information about Turing reducibility.

We characterize the ordinals α of uncountable cofinality such that α is the standard part of a nonstandard model of ZFC (or equivalently KP).

Le paradigme de théorie stable est la théorie T d'un corps algébriquement clos; une autre théorie stable T′ est celle de la structure formée d'un corps algébriquement clos, avec en outre un symbole relationnel unaire interprétant un de ses sous-corps propres algébriquement clos. C'est à l'éclaircissement des rapports de T et de T′ qu'est consacré cet article.

J'y considère une théorie complète T stable, et les structures formées d'un modèle N de T, avec en outre un symbole relationnel unaire (x) interprétant une restriction élémentaire M de N; j'appelle ces structures paires de modèles de T.

Et je dis que la paire (N, M) est belle si d'une part M est ∣T∣+-saturé, et d'autre part pour tout n-uplet ā d'éléments de N, tout type, au sens de T, sur M ⋃ {α} est réalisé dans N.

Le premier résultat (Théorème 4) est que deux belles paires sont élémentairement équivalentes. Plus précisément, si (N1, M1) et (N2, M2) sont deux belles paires, et si ā est dans la première, b¯ dans la seconde, le fait que le type de ā sur M1 et celui de b¯ sur M2 soient équivalents dans l'ordre fondamental au sens de T suffit (et est bien sûr nécessaire) pour que ā at b¯ aient même type (sur ⊘) au sens de la théorie T′ des belles paires.

We prove that the logics of Magidor-Malitz and their generalization by Rubin are distinct even for PC classes.

Let M ⊨ Qnx1 … xnφ(x1 … xn) mean that there is an uncountable subset A of ∣M∣ such that for every a1 …, an ∈ A, M ⊨ φ[a1, …, an].

Theorem 1.1 (Shelah) (♢ℵ1). For every n ∈ ωthe classKn+1 = {‹A, R› ∣ ‹A, R› ⊨ ¬ Qn+1x1 … xn+1R(x1, …, xn+1)} is not an ℵ0-PC-class in the logic ℒn, obtained by closing first order logic underQ1, …, Qn. I.e. for no countable ℒn-theory T, isKn+1the class of reducts of the models of T.

Theorem 1.2 (Rubin) (♢ℵ1). Let M ⊨ QE x yφ(x, y) mean that there is A ⊆ ∣M∣ such thatEA, φ = {‹a, b› ∣ a, b ∈ A and M ⊨ φ[a, b]) is an equivalence relation on A with uncountably many equivalence classes, and such that each equivalence class is uncountable. Let KE = {‹A, R› ∣ ‹A, R› ⊨ ¬ QExyR(x, y)}. Then KE is not an ℵ0-PC-class in the logic gotten by closing first order logic under the set of quantifiers {Qn ∣ n ∈ ω) which were defined in Theorem 1.1.

In [6, p. 317] Curry described a formal system assigning types to terms of the type-free λ-calculus. In [11] Scott gave a natural semantics for this type assignment and asked whether a completeness result holds.

Inspired by [4] and [5] we extend the syntax and semantics of the Curry types in such a way that filters in the resulting type structure form a domain in the sense of Scott [12]. We will show that it is possible to turn the domain of types into a λ-model, among other reasons because all λ-terms possess a type. This model gives the completeness result for the extended system. By a conservativity result the completeness for Curry's system follows.

Independently Hindley [8], [9] has proved both completeness results using term models. His method of proof is in some sense dual to ours.

For λ-calculus notation see [1].

The set Λ of isols was extensively studied by Dekker and Myhill in [1]. Subsequently, Nerode [3] developed the theory of Λ(A), the set of isols relative to some recursively closed set of functions A.

One of the main areas of interest of [1] was the natural partial order ≤ on Λ. In this paper we will examine some of the properties of ≤A on Λ(A). We use the following notations: ∣A∣ is the cardinality of the set A, ⊃ denotes strict inclusion, (a) is the power set of the set a, c is the cardinality of the continuum, and ω = {0, 1, 2, …}. The terms A-isol, A-immune, A-r.e., A-incomparable, etc. all refer to the usual meaning of these words, only taken in the context of the recursively closed set A. ReqA(a) is the A-r.e.t. of which a is a representative. By identifying a finite natural number with the A-r.e.t. consisting of sets of a given finite cardinality we see that ω ⊆ Λ(A); Λ(A) is said to be nontrivial iff ω ⊃ Λ(A). The three results proven in this paper are:

1. Theorem 1. If Λ(A) is nontrivial, then ∣Λ(A)∣ = c.

2. Theorem 2. If∣A∣ < c, then Λ(A) is nontrivial.

3. Theorem 3. If ∣A∣ < c and ∣⊿∣ < c and ⊿ ⊆ Λ(A) − ω, then there is aΓ ⊆ Λ(A) − ω such that:

• (a) ∣Γ∣ = c.

• (b) Every member of Γ is A-incomparable with every member of Δ.

• (c) Any two distinct members of Γ are A-incomparable.

Let ZF denote Zermelo-Fraenkel set theory (without the axiom of choice), and let M be a countable transitive model of ZF. The method of forcing extends M to another model M[G] of ZF (a “generic extension”). If the axiom of choice holds in M it also holds in M[G], that is, the axiom of choice is preserved by generic extensions. We show that this is not true for many weak forms of the axiom of choice, and we derive an application to Boolean toposes.

In this paper some partial models for combinators with weak reduction are proposed. The underlying ideas are due to Church [1]. The results are extensions of the previous results of the author proved in [8].

A well-known result, going back to the twenties, states that, under some reasonable assumptions, any logic can be characterized as the set of formulas satisfied by a matrix 〈, F〉, where is an algebra of the appropriate type, and F a subset of the domain of , called the set of designated elements. In particular, every quasi-classical modal logic—a set of modal formulas, containing the smallest classical modal logic E, which is closed under the inference rules of substitution and modus ponens—is characterized by such a matrix, where now is a modal algebra, and F is a filter of . If the modal logic is in fact normal, then we can do away with the filter; we can study normal modal logics in the setting of varieties of modal algebras. This point of view was adopted already quite explicitly in McKinsey and Tarski [8]. The observation that the lattice of normal modal logics is dually isomorphic to the lattice of subvarieties of a variety of modal algebras paved the road for an algebraic study of normal modal logics. The algebraic approach made available some general results from Universal Algebra, notably those obtained by Jónsson [6], and thereby was able to contribute new insights in the realm of normal modal logics [2], [3], [4], [10].

The requirement that a modal logic be normal is rather a severe one, however, and many of the systems which have been considered in the literature do not meet it. For instance, of the five celebrated modal systems, S1–S5, introduced by Lewis, S4 and S5 are the only normal ones, while only SI fails to be quasi-classical. The purpose of this paper is to generalize the algebraic approach so as to be applicable not just to normal modal logics, but to quasi-classical modal logics in general.

We start with the framework upon which this paper is based. The most useful reference for these notions is [2]. For any nonempty index set I and any proper filter D on S(I) (the power set of I), we denote by I/D the reduced power of modulo D as defined in [2, pp. 167–169]. The first-order language associated with I/D will always be the same language as associated with . We denote the 2-element Boolean algebra 〈{0, 1}, ⋂, ⋃, c, 0, 1〉 by 2 and 2I/D denotes the reduced power of it modulo D. We point out the intimate connection between the structures I/D and 2I/D given in [2, pp. 341–345]. Moreover, we assume as known the definition of Horn formula and Horn sentence as given in [2, p. 328] along with the fundamental theorem that φ is a reduced product sentence iff φ is provably equivalent to a Horn sentence [2, Theorem 6.2.5/ (iff φ is a 2-direct product sentence and a reduced power sentence [2, Proposition 6.2.6(ii)]). For a theory T(any set of sentences), ⊨ T denotes that is a model of T.

In addition to the above we assume as known the elementary characteristics (due to Tarski) associated with a complete theory of a Boolean algebra, and we adopt the notation 〈n, p, q〉 of [3], [10], or [6] to denote such an elementary characteristic or the corresponding complete theory. We frequently will use Ershov's theorem which asserts that for each 〈n, p, q〉 there exist an index set I and filter D such that 2I/D ⊨ 〈n, p, q〉 [3] or [2, Lemma 6.3.21].

We study cofinal extensions of models of arithmetic, in particular we show that some properties near to expandability are preserved under cofinal extensions.

The results in this paper were motivated by the following result due to R. Solovay.

Theorem 1 (Solovay). Let M be a nonstandard model of Peano's first order axioms P and let I ⊂e M (i.e. ϕ ≠ ⊂ M and I is closed under < and successor). Then for each of the functions we can define J ⊆e I in ‹M, I› such that J is closed under that function. (∣x∣ denotes [log2(x)].)

Proof. Just notice that the cuts defined by

are successively closed under

In view of Theorem 1, the following question was raised by R. Solovay: Can we define J ⊆ I in ‹M, I› such that J is closed under exponentiation? In Theorem 2 we show that the answer is “no”. Theorem 3 is based on Theorem 2 and extends the technique to cuts which are models of subsystems of P.

To prove both theorems we shall need an estimate due to R. Parikh (see [1], especially the proof of Theorem 2.2a). For the sake of completeness, and also to introduce some notation we shall sketch Parikh's estimate in the next section. At all times we shall give the easiest estimates which still work rather than the sharpest ones.

One of the objectives of mathematics is to construct suitable models for practical or theoretical phenomena and to explore the mathematical richness of such models. This enables other scientists to obtain a better understanding of such phenomena. As an example we will mention the real line and related structures. The line can be used profitably in the study of discrete phenomena like population growth, chemical reactions, etc.

Today's version of the real line is a topological completion of the rational numbers. This is so because then mathematicians have been able to work out a powerful analysis of the line. By using the real line to construct models for finitary phenomena we are more able to study those phenomena than we would have been sticking only to true-to-nature but finite structures.

So we may say that the line is a mathematical model for certain finite structures. This motivates us to seek natural models for other types of finite structures, and it is natural to look for models that in some sense are complete.

In this paper our starting point will be finite systems of finite operators. For the sake of simplicity we assume that they all are operators of one variable and that all the values are natural numbers. There is a natural extension of the systems such that they accept several variables and give finite operators as values, but the notational complexity will then obscure the idea of the construction.