Ω-consistent theory


In mathematical logic, an ω-consistent theory is a theory that is not only consistent, but also avoids proving certain infinite combinations of sentences that are intuitively contradictory. The name is due to Kurt Gödel, who introduced the concept in the course of proving the incompleteness theorem.

Definition

A theory T is said to interpret the language of arithmetic if there is a translation of formulas of arithmetic into the language of T so that T is able to prove the basic axioms of the natural numbers under this translation.
A T that interprets arithmetic is ω-inconsistent if, for some property P of natural numbers, T proves P, P, P, and so on, but T also proves that there is some natural number n such that P fails. This may not generate a contradiction within T because T may not be able to prove for any specific value of n that P fails, only that there is such an n.
T is ω-consistent if it is not ω-inconsistent.
There is a weaker but closely related property of Σ1-soundness. A theory T is Σ1-sound if every Σ01-sentence provable in T is true in the standard model of arithmetic N.
If T is strong enough to formalize a reasonable model of computation, Σ1-soundness is equivalent to demanding that whenever T proves that a Turing machine C halts, then C actually halts. Every ω-consistent theory is Σ1-sound, but not vice versa.
More generally, we can define an analogous concept for higher levels of the arithmetical hierarchy. If Γ is a set of arithmetical sentences, a theory T is Γ-sound if every Γ-sentence provable in T is true in the standard model. When Γ is the set of all arithmetical formulas, Γ-soundness is called just soundness.
If the language of T consists only of the language of arithmetic, then a sound system is one whose model can be thought of as the set ω, the usual set of mathematical natural numbers. The case of general T is different, see [|ω-logic] below.
Σn-soundness has the following computational interpretation: if the theory proves that a program C using a Σn−1-oracle halts, then C actually halts.

Examples

Consistent, ω-inconsistent theories

Write PA for the theory Peano arithmetic, and Con for the statement of arithmetic that formalizes the claim "PA is consistent". Con could be of the form "For every natural number n, n is not the Gödel number of a proof from PA that 0=1".
Now, assuming PA is really consistent, it follows that PA + ¬Con is also consistent, for if it were not, then PA would prove Con, contradicting Gödel's second incompleteness theorem. However, PA + ¬Con is not ω-consistent. This is because, for any particular natural number n, PA + ¬Con proves that n is not the Gödel number of a proof that 0=1. However, PA + ¬Con proves that, for some natural number n, n is the Gödel number of such a proof.
In this example, the axiom ¬Con is Σ1, hence the system PA + ¬Con is in fact Σ1-unsound, not just ω-inconsistent.

Arithmetically sound, ω-inconsistent theories

Let T be PA together with the axioms cn for each natural number n, where c is a new constant added to the language. Then T is arithmetically sound, but ω-inconsistent.
Σ1-sound ω-inconsistent theories using only the language of arithmetic can be constructed as follows. Let IΣn be the subtheory of PA with the induction schema restricted to Σn-formulas, for any n > 0. The theory IΣn + 1 is finitely axiomatizable, let thus A be its single axiom, and consider the theory T = IΣn + ¬A. We can assume that A is an instance of the induction schema, which has the form
If we denote the formula
by P, then for every natural number n, the theory T proves P. On the other hand, T proves the formula, because it is logically equivalent to the axiom ¬A. Therefore, T is ω-inconsistent.
It is possible to show that T is Πn + 3-sound. In fact, it is Πn + 3-conservative over the theory IΣn. The argument is more complicated.

Arithmetically unsound, ω-consistent theories

Let ω-Con be the arithmetical sentence formalizing the statement "PA is ω-consistent". Then the theory PA + ¬ω-Con is unsound, but ω-consistent. The argument is similar to the first example: a suitable version of the Hilbert-Bernays-Löb derivability conditions holds for the "provability predicate" ω-Prov = ¬ω-Con, hence it satisfies an analogue of Gödel's second incompleteness theorem.

ω-logic

The concept of theories of arithmetic whose integers are the true mathematical integers is captured by ω-logic. Let T be a theory in a countable language which includes a unary predicate symbol N intended to hold just of the natural numbers, as well as specified names 0, 1, 2,..., one for each natural number. Note that T itself could be referring to more general objects, such as real numbers or sets; thus in a model of T the objects satisfying N are those that T interprets as natural numbers, not all of which need be named by one of the specified names.
The system of ω-logic includes all axioms and rules of the usual first-order predicate logic, together with, for each T-formula P with a specified free variable x, an infinitary ω-rule of the form:
That is, if the theory asserts P separately for each natural number n given by its specified name, then it also asserts P collectively for all natural numbers at once via the evident finite universally quantified counterpart of the infinitely many antecedents of the rule. For a theory of arithmetic, meaning one with intended domain the natural numbers such as Peano arithmetic, the predicate N is redundant and may be omitted from the language, with the consequent of the rule for each P simplifying to.
An ω-model of T is a model of T whose domain includes the natural numbers and whose specified names and symbol N are standardly interpreted, respectively as those numbers and the predicate having just those numbers as its domain. If N is absent from the language then what would have been the domain of N is required to be that of the model, i.e. the model contains only the natural numbers. These requirements make the ω-rule sound in every ω-model. As a corollary to the omitting types theorem, the converse also holds: the theory T has an ω-model if and only if it is consistent in ω-logic.
There is a close connection of ω-logic to ω-consistency. A theory consistent in ω-logic is also ω-consistent. The converse is false, as consistency in ω-logic is a much stronger notion than ω-consistency. However, the following characterization holds: a theory is ω-consistent if and only if its closure under unnested applications of the ω-rule is consistent.

Relation to other consistency principles

If the theory T is recursively axiomatizable, ω-consistency has the following characterization, due to Craig Smoryński:
Here, is the set of all Π02-sentences valid in the standard model of arithmetic, and is the uniform reflection principle for T, which consists of the axioms
for every formula with one free variable. In particular, a finitely axiomatizable theory T in the language of arithmetic is ω-consistent if and only if T + PA is -sound.