Results in recursion-theoretic inductive inference have been criticized as depending on unrealistic self-referential examples. J. M. Bārzdiņš proposed a way of ruling out such examples, and conjectured that one of the earliest results of inductive inference theory would fall if his method were used. In this paper we refute Bārzdiņš' conjecture.

We propose a new line of research examining robust separations; these are defined using a strengthening of Bārzdiņš' original idea. The preliminary results of the new line of research are presented, and the most important open problem is stated as a conjecture. Finally, we discuss the extension of this work from function learning to formal language learning.

The constructible universe L of Gödel [2] has a natural well-ordering < L given by the order of construction; a closer look reveals that this well-ordering is definable by a Σ1 formula. Cohen's method of forcing provides several examples of models of ZF + V ≠ L which have a definable well-ordering but none is definable by a relatively simple formula.

Recently, Mansfield [7] has shown that if a set of reals (or hereditarily countable sets) has a Σ1, well-ordering then each of its elements is constructible. A question has thus arisen whether one can find a model of ZF + V ≠ L that has a Σ1 well-ordering of the universe. We answer this question in the affirmative.

The main result of this paper is

Theorem. There is a model of ZF + V ≠ L which has a Σ1 well-ordering.

The model is a generic extension of L by adjoining a branch through a Suslin tree with certain properties. The branch is a nonconstructible subset of ℵ1. Note that by Mansfield's theorem, the model must not have nonconstructible subsets of ω.

Our results can be generalized in several directions. We note that in particular, we can get a model with a Σ1 well-ordering that is not L[X] for any set X. As one might expect from a joint paper by a recursion theorist and a set theorist, the proof consists of a construction and a computation.

In [3], Shelah defined the cardinals κn(T) and , for each theory T and n < ω. κn(T) is the least cardinal κ without a sequence (pi)i<κ of complete n-types such that pi is a forking extension of pj for all i < j < κ. It is essential in computing the stability spectrum of a stable theory. On the other hand is called the number of independent partitions of T. (See Definition 1.2 below.) Unfortunately this invariant has not been investigated deeply. In the author's opinion, this unfortunate situation of is partially due to the fact that its definition is complicated in expression. In this paper, we shall give equivalents of which can be easily handled.

In §1 we shall state the definitions of κn(T) and . Some basic properties of forking will be stated in this section. We shall also show that if = ∞ then T has the independence property.

In §2 we shall give some conditions on κ, n, and T which are equivalent to the statement . (See Theorem 2.1 below.) We shall show that does not depend on n. We introduce the cardinal ı(T), which is essential in computing the number of types over a set which is independent over some set, and show that ı(T) is closely related to . (See Theorems 2.5 and 2.6 below.) The author expects the reader will discover the importance of via these theorems.

Some of our results are motivated by exercises and questions in [3, Chapter III, §7]. The author wishes to express his heartfelt thanks to the referee for a number of helpful suggestions.

In [5] Scott asked if there was a proof theoretic proof of his interpolation theorem. The purpose of this paper is to provide such a proof, working with the first order system LK of Gentzen [2]. Our method is an extension of the one in Maehara [3] for Craig's interpolation theorem. We will also sketch the original model theoretic proof and show how Scott used his result to obtain a definability theorem of Svenonius [7].

A language for LK contains the usual logical symbols: , ∧, ∨, ⊃, ∀, ∃; countably many free variables a0, a1, … and bound variables x0, x1, …; and some or all of the following nonlogical symbols: n-ary predicates ; n-ary functions ; and individual constants c0, c1, …. Semiterms are defined as follows: (1) Free variables, bound variables and individual constants are semiterms. (2) If f is an n-ary function and s1 …, sn are semiterms, then f(s1 …, sn), is a semiterm. A term is a semiterm that does not contain a bound variable. Formulas are defined as follows: (1) If R is an n-ary predicate and t1 …, tn are terms, then R(t1 …, tn) is a formula. (2) If A and B are formulas, then A, A ∧ B, A ∨ B and A ⊃ B are formulas. (3) If A(t) is a formula which has zero or more occurrences of the term t, and if x is a bound variable not contained in A(t), then ∀xA(x) and ∃xA(x) are formulas where A(x) is obtained from A(t) by substituting x for t at all indicated places.

In this paper a converse of Barwise's completeness theorem is proved by cut-elimination considerations applied to inductive definitions. We show that among the transitive sets T satisfying some weak closure conditions (closure under primitive-recursive set-functions is more than enough), only the unions of admissible sets satisfy Barwise's completeness theorem in the form stating that if φ ∊ T is a sentence which has a derivation (in the universe) then φ has a derivation in T. See §1 for the origin of the problem in Barwise's paper [Ba].

Stated quite briefly the proof is as follows (a step-by-step account including relevant definitions is given in the body of the paper):

Let T be a transitive prim.-rec. closed set, and let is nonempty, transitive and closed under pairs}. For each let κ(A) be the supremum of closure ordinals of first-order positive operators on subsets of A (first-order with respect to By Theorem 1 of [BGM], it is enough to prove that rank(T) in order to obtain that T is a union of admissible sets. (The rank of a set is defined by rank(x) = supy ∊ x (rank(y) + 1); since T is prim.-rec. closed, rank(T) = smallest ordinal not in T.)

Let We show how to find in T (in fact, in Lω(A)) a derivable sentence τ that has no derivation D such that rank(D) ≤ α. Thus, if τ is to have a derivation in T, rank(T) > α. α is arbitrary (< κ(A)), so rank(T) ≥ κ(A). Q.E.D.

In this note, we consider Keisler's stability theory and prove that every measure over a small submodel has a locally pure extension.

We define, in the spirit of Fenstad [2], a higher type computation theory, and show that countable recursion over the continuous functionals forms such a theory. We also discuss Hyland's proposal from [4] for a scheme with which to supplement S1–S9, and show that this augmented set of schemes fails to generate countable recursion. We make another proposal to which the methods of this section do not apply.

Myhill [12] extended the ideas of Shapiro [15], and proposed a system of epistemic set theory IST (based on modal S4 logic) in which the meaning of the necessity operator is taken to be the intuitive provability, as formalized in the system itself. In this setting one works in classical logic, and yet it is possible to make distinctions usually associated with intuitionism, e.g. a constructive existential quantifier can be expressed as (∃x) □ …. This was first confirmed when Goodman [7] proved that Shapiro's epistemic first order arithmetic is conservative over intuitionistic first order arithmetic via an extension of Gödel's modal interpretation [6] of intuitionistic logic.

Myhill showed that whenever a sentence □A ∨ □B is provable in IST, then A is provable in IST or B is provable in IST (the disjunction property), and that whenever a sentence ∃x.□A(x) is provable in IST, then so is A(t) for some closed term t (the existence property). He adapted the Friedman slash [4] to epistemic systems.

Goodman [8] used Epistemic Replacement to formulate a ZF-like strengthening of IST, and proved that it was a conservative extension of ZF and that it had the disjunction and existence properties. It was then shown in [13] that a slight extension of Goodman's system with the Epistemic Foundation (ZFER, cf. §1) suffices to interpret intuitionistic ZF set theory with Replacement (ZFIR, [10]). This is obtained by extending Gödel's modal interpretation [6] of intuitionistic logic. ZFER still had the properties of Goodman's system mentioned above.

Let A be a non-empty set. A set is said to be stationary in if for every f: [A]<ω → A there exists x ϵ S such that x ≠ A and f“[x]<ω ⊆ x. In this paper we prove the following: For an uncountable cardinal λ and a stationary set S in , if there is a regular uncountable cardinal κ ≤ λ such that {x ϵ S: x ∩ κ ϵ κ} is stationary, then S can be split into κ disjoint stationary subsets.

We present three examples of countable homogeneous structures (also called Fraïssé limits) whose automorphism groups are not universal, namely, fail to contain isomorphic copies of all automorphism groups of their substructures.

Our first example is a particular case of a rather general construction on Fraïssé classes, which we call diversification, leading to automorphism groups containing copies of all finite groups. Our second example is a special case of another general construction on Fraïssé classes, the mixed sums, leading to a Fraïssé class with all finite symmetric groups appearing as automorphism groups and at the same time with a torsion-free automorphism group of its Fraïssé limit. Our last example is a Fraïssé class of finite models with arbitrarily large finite abelian automorphism groups, such that the automorphism group of its Fraïssé limit is again torsion-free.

Let be the least cardinal θ such that MAθ fails, (i.e. MA) implies that is regular. Models for regular with are easy to get (see [Ku1]). Fremlin and Miller proved that cof() > ω [Fr, 41C(d)]. The question of whether it is consistent that be singular was solved, in the affirmative, by Kunen [Ku1]. Kunen used a (θ, θ) strong gap in (ω)/Fin, whose splitting partial order is c.c.c. He showed that if cof(θ) = ω1, and P is a c.c.c. partial order of cardinality < θ that may destroy the strong gap, then there exists another c.c.c. partial order Q, which does not destroy the strong gap, and such that ⊩Q “P is not c.c.c”. One then gets Kunen's model by iterating c.c.c. partial orders of cardinality < θ, without destroying the strong (θ, θ) gap. It is unknown whether it is consistent to have ω1 < cof() < .

Clearly, there exists a c.c.c. partial order Q with ∣Q∣ = , such that (Q) = (see (1.1)). A partial order P is σ-centered iff P is the union of countably many centered subsets, where a subset A ⊆ P is centered iff p ≤ q. Clearly, a σ-centered partial order is c.c.c. Bell and Szymański proved that (σ-centered) is regular (see [Be] or [Fr, 14C], and [Fr, 21K] or [vD, 3.1(e)]). This implies that if P is a σ-centered partial order, and is singular, then (P) > . In particular, if is singular, then Q is not σ-centered and all c.c.c. partial orders of cardinality < have a Baire number (1.1) strictly greater than . It was suggested in [Ku1] to try and use Q(T) (1.2), where T is a tree with no ω1-branches, to get models with singular . It is well known that when T is such a tree, the forcing Q(T) is c.c.c. [BMR], ∣Q(T)∣ = ∣T∣, and if T is not special, then n(Q(T)) ≤ ∣T∣.

This paper continues the project initiated in [5]: a model-theoretic study of the concept of cardinality within certain higher-order logics. As recommended by an editor of this Journal, I will digress to say something about the project's motivation. Then I will review some of the basic definitions from [5]; for unexplained notation the reader should consult [5].

The syntax of ordinary usage (with respect to the construction of arguments as well as the construction of individual sentences) makes it natural to classify numerals and expressions of the form ‘the number of F's’ as singular terms, expressions like ‘is prime’ or ‘is divisible by’ as predicates of what Frege called “level one”, and expressions like ‘for some natural number’ as first-order quantifier-phrases. From this syntatic classification, it is a short step—so short as to be frequently unnoticed—to a semantic thesis: that such expressions play the same sort of semantic role as is played by the paradigmatic (and nonmathematical) members of these lexical classes. Thus expressions of the first sort are supposed to designate objects (in post-Fregean terms, entities of type 0), those of the second sort to be true or false of tuples of objects, and those of the third sort to quantify over objects. All this may be summed up in Frege's dictum: “Numbers are objects.”