The second axiomatization of set theory (see the Click Here to see full-size tabletable of Neumann-Bernays-Gödel axioms) originated with John von Neumann in the 1920s. His formulation differed considerably from ZFC because the notion of function, rather than that of set, was taken as undefined, or “primitive.” In a series of papers beginning in 1937, however, the Swiss logician Paul Bernays, a collaborator with the German formalist David Hilbert, modified the von Neumann approach in a way that put it in much closer contact with ZFC. In 1940, the Austrian-born American logician Kurt Gödel, known for his undecidability proof, further simplified the theory. This axiomatic version of set theory is called NBG, after the Neumann-Bernays-Gödel axioms. As will be explained shortly, NBG is closely related to ZFC, but it allows explicit treatment of so-called classes: collections that might be too large to be sets, such as the class of all sets or the class of all ordinal numbers.

For expository purposes it is convenient to adopt two undefined notions for NBG: class and the binary relation ∊ of membership (though, as is also true in ZFC, ∊ suffices). For the intended interpretation, variables take classes—the totalities corresponding to certain properties—as values. A class is defined to be a set if it is a member of some class; those classes that are not sets are called proper classes. Intuitively, sets are intended to be those classes that are adequate for mathematics, and proper classes are thought of as those collections that are “so big” that, if they were permitted to be sets, contradictions would follow. In NBG, the classical paradoxes are avoided by proving in each case that the collection on which the paradox is based is a proper class—i.e., is not a set.

Comments about the axioms that follow are limited to features that distinguish them from their counterpart in ZFC. The axiom schema for class formation is presented in a form to facilitate a comparison with the axiom schema of separation of ZFC. In a detailed development of NBG, however, there appears instead a list of seven axioms (not schemas) that state that, for each of certain conditions, there exists a corresponding class of all those sets satisfying the condition. From this finite set of axioms, each an instance of the above schema, the schema (in a generalized form) can be obtained as a theorem. When obtained in this way, the axiom schema for class formation of NBG is called the class existence theorem.

In brief, axioms 4 through 8 in the table of NBG are axioms of set existence. The same is true of the next axiom, which for technical reasons is usually phrased in a more general form. Finally, there may appear in a formulation of NBG an analog of the last axiom of ZFC (axiom of restriction).

A comparison of the two theories that have been formulated is in order. In contrast to the axiom schema of replacement of ZFC, the NBG version (axiom 9 in the table) is not an axiom schema but an axiom. Thus, with the comments above about the ZFC axiom schema of separation in mind, it follows that NBG has only a finite number of axioms. On the other hand, since the axiom schema of replacement of ZFC provides an axiom for each formula, ZFC has infinitely many axioms—which is unavoidable because it is known that no finite subset yields the full system of axioms. The finiteness of the axioms for NBG makes the logical study of the system simpler. The relationship between the theories may be summarized by the statement that ZFC is essentially the part of NBG that refers only to sets. Indeed, it has been proved that every theorem of ZFC is a theorem of NBG and that any theorem of NBG that speaks only about sets is a theorem of ZFC. From this it follows that ZFC is consistent if and only if NBG is consistent.

Limitations of axiomatic set theory

The fact that NBG avoids the classical paradoxes and that there is no apparent way to derive any one of them in ZFC does not settle the question of the consistency of either theory. One method for establishing the consistency of an axiomatic theory is to give a model—i.e., an interpretation of the undefined terms in another theory such that the axioms become theorems of the other theory. If this other theory is consistent, then that under investigation must be consistent. Such consistency proofs are thus relative: the theory for which a model is given is consistent if that from which the model is taken is consistent. The method of models, however, offers no hope for proving the consistency of an axiomatic theory of sets. In the case of set theory and, indeed, of axiomatic theories generally, the alternative is a direct approach to the problem.

If T is the theory of which the (absolute) consistency is under investigation, this alternative means that the proposition “There is no sentence of T such that both it and its negation are theorems of T” must be proved. The mathematical theory (developed by the formalists) to cope with proofs about an axiomatic theory T is called proof theory, or metamathematics. It is premised upon the formulation of T as a formal axiomatic theory—i.e., the theory of inference (as well as T) must be axiomatized. It is then possible to present T in a purely symbolic form—i.e., as a formal language based on an alphabet the symbols of which are those for the undefined terms of T and those for the logical operators and connectives. A sentence in this language is a formula composed from the alphabet according to prescribed rules. The hope for metamathematics was that, by using only intuitively convincing, weak number-theoretic arguments (called finitary methods), unimpeachable proofs of the consistency of such theories as axiomatic set theory could be given.

That hope suffered a severe blow in 1931 from a theorem proved by Kurt Gödel about any formal theory S that includes the usual vocabulary of elementary arithmetic. By coding the formulas of such a theory with natural numbers (now called Gödel numbers) and by talking about these numbers, Gödel was able to make the metamathematics of S become part of the arithmetic of S and hence expressible in S. The theorem in question asserts that the formula of S that expresses (via a coding) “S is consistent” in S is unprovable in S if S is consistent. Thus, if S is consistent, then the consistency of S cannot be proved within S; rather, methods beyond those that can be expressed or reflected in S must be employed. Because, in both ZFC and NBG, elementary arithmetic can be developed, Gödel’s theorem applies to these two theories. Although there remains the theoretical possibility of a finitary proof of consistency that cannot be reflected in the foregoing systems of set theory, no hopeful, positive results have been obtained.

Other theorems of Gödel when applied to ZFC (and there are corresponding results for NBG) assert that, if the system is consistent, then (1) it contains a sentence such that neither it nor its negation is provable (such a sentence is called undecidable), (2) there is no algorithm (or iterative process) for deciding whether a sentence of ZFC is a theorem, and (3) these same statements hold for any consistent theory resulting from ZFC by the adjunction of further axioms or axiom schemas. Apparently ZFC can serve as a foundation for all of present-day mathematics because every mathematical theorem can be translated into and proved within ZFC or within extensions obtained by adding suitable axioms. Thus, the existence of undecidable sentences in each such theory points out an inevitable gap between the sentences that are true in mathematics and sentences that are provable within a single axiomatic theory. The fact that there is more to conceivable mathematics than can be captured by the axiomatic approach prompted the American logician Emil Post to comment in 1944 that “mathematical thinking is, and must remain, essentially creative.”