# Von Neumann–Bernays–Gödel set theory

In the foundations of mathematics, von Neumann–Bernays–Gödel set theory (NBG) is an axiomatic set theory that is a conservative extension of the canonical Zermelo–Fraenkel set theory (ZFC). A statement in the language of ZFC is provable in NBG if and only if it is provable in ZFC. The ontology of NBG includes proper classes, objects having members but that cannot be members of other entities. NBG's principle of class comprehension is predicative; quantified variables in the defining formula can range only over sets. Allowing impredicative comprehension turns NBG into Morse-Kelley set theory (MK). NBG, unlike ZFC and MK, can be finitely axiomatized.

## Ontology

The defining aspect of NBG is the distinction between proper class and set. Let a and s be two individuals. Then the atomic sentence ${\displaystyle a\in s}$ is defined if a is a set and s is a class. In other words, ${\displaystyle a\in s}$ is defined unless a is a proper class. A proper class is very large; NBG even admits of "the class of all sets", the universal class called V. However, NBG does not admit "the class of all classes" (which fails because proper classes are not "objects" that can be put into classes in NBG) or "the set of all sets" (whose existence cannot be justified with NBG axioms).

By NBG's axiom schema of Class Comprehension, all objects satisfying any given formula in the first-order language of NBG form a class; if a class is not a set in ZFC, it is an NBG proper class.

The development of classes mirrors the development of naive set theory. The principle of abstraction is given, and thus classes can be formed out of all individuals satisfying any statement of first-order logic whose atomic sentences all involve either the membership relation or predicates definable from membership. Equality, pairing, subclass, and such, are all definable and so need not be axiomatized — their definitions denote a particular abstraction of a formula.

Sets are developed in a manner very similarly to ZF. Let Rp(A,a), meaning "the set a represents the class A," denote a binary relation defined as follows:

${\displaystyle \mathrm {Rp} (A,a):=\forall x(x\in A\leftrightarrow x\in a).}$

That is, a "represents" A if every element of a is an element of A, and conversely. Classes lacking representations, such as the class of all sets that do not contain themselves (the class invoked by the Russell paradox), are the proper classes.

## History

In two articles published in 1925 and 1928, John von Neumann stated his axioms and showed they were adequate to develop set theory.[1] Von Neumann took functions and arguments as primitives. His functions correspond to classes, and functions that can be used as arguments correspond to sets. In fact, he defined classes and sets using functions that can take only two values (that is, indicator functions whose domain is the class of all arguments).

Von Neumann's work in set theory was influenced by Cantor's articles, Zermelo's 1908 axioms for set theory, and the 1922 critiques of Zermelo's set theory that were given independently by Fraenkel and Skolem. Both Fraenkel and Skolem pointed out that Zermelo's axioms cannot prove the existence of the set {Z0, Z1, Z2, … } where Z0 is the set of natural numbers, and Zn+1 is the power set of Zn. They then introduced the axiom of replacement, which would guarantee the existence of such sets.[2] However, they were reluctant to adopt this axiom: Fraenkel's opinion was "that Replacement was too strong an axiom for 'general set theory' … and … Skolem only wrote that 'we could introduce' Replacement".[3]

Von Neumann worked on the deficiencies in Zermelo's set theory and introduced several innovations to remedy them, including:

• A theory of ordinals. Zermelo's set theory does not contain Cantor's theory of ordinal numbers. Von Neumann recovered this theory by defining the ordinals using sets that are well-ordered by the ∈-relation.[4] In contrast to Fraenkel and Skolem, von Neumann found the axiom of replacement so essential to his work that he declared: "In fact, I believe that no theory of ordinals is possible at all without this axiom."[5]
• A criterion identifying classes that are too large to be sets. Zermelo did not provide such a criterion. His set theory avoids the large classes that lead to the paradoxes, but it leaves out many sets, such as the one mentioned by Fraenkel and Skolem.[6] Von Neumann's criterion is: A class is too large to be a set if and only if it can be mapped onto the universal class. Von Neumann realized that the paradoxes can be avoided by not allowing such large classes to be members of any class. Combining this restriction with his criterion, he obtained his axiom of limitation of size: A class X is not a member of any class if and only if X can be mapped onto the universal class.[7] He proved that this axiom implies the axioms of replacement and separation, and implies that the universal class can be well-ordered (which is equivalent to the axiom of global choice).
• Finite axiomatization. Fraenkel and Skolem formalized Zermelo's imprecise concept of "definite propositional function", which appears in his axiom of separation. Skolem gave the axiom schema of separation that is currently used in ZFC; Fraenkel gave an equivalent approach. Zermelo rejected both approaches "particularly because they implicitly involve the concept of natural number which, in Zermelo's view, should be based upon set theory."[8] Von Neumann avoided axiom schemas by formalizing the concept of "definite propositional function" with his functions, whose construction requires only finitely many axioms. This led to his set theory having finitely many axioms.[9] (In 1961, Montague proved that ZFC cannot be finitely axiomatized.[10])
• The axiom of regularity. Zermelo's set theory does not exclude non-well-founded sets.[11] Fraenkel and von Neumann introduced axioms to exclude these sets. Von Neumann introduced the axiom of regularity, which states that all sets are well-founded.[12] Although von Neumann did not adopt regularity as an axiom,[13] he proved its relative consistency while studying his axiom of limitation of size. First he weakened his axiom system by replacing the latter axiom with two of its consequences: replacement and a choice axiom equivalent to global choice.[14] Next he proved that if this weaker system is consistent, it remains consistent after adding the axiom of regularity. Finally, he showed that his weaker system augmented with regularity proves the axiom of limitation of size.[15] These results establish that the axioms of regularity and limitation of size are relatively consistent with respect to his weaker system, and that (in the presence of regularity and his other axioms) replacement and his choice axiom are equivalent to the axiom of limitation of size.

In a series of articles published between 1937 and 1954, Paul Bernays modified von Neumann's theory by taking sets and classes as primitives. By using sets, Bernays was following the tradition established by Cantor, Dedekind, and Zermelo. His classes followed the tradition of Boolean algebra since they permit the operation of complement as well as union and intersection.[16] Bernays handled sets and classes in a two-sorted logic. This required the introduction of two membership primitives: one for membership in sets, and one for membership in classes. With these primitives, Bernays rewrote and simplified von Neumann's axioms. He also adopted the axiom of regularity, and replaced the axiom of limitation of size with the axioms of replacement and von Neumann's choice axiom. (Von Neumann's work shows that the last two changes allow Bernays' axioms to prove the axiom of limitation of size.) [17]

Kurt Gödel simplified Bernays' theory by making every set a class, which allowed him to use just one sort for classes and one membership primitive. He also introduced a predicate indicating which classes are sets. Gödel modified some of Bernays' axioms, and introduced the axiom of global choice to replace von Neumann's choice axiom. He used his axioms in his 1940 monograph on the relative consistency of global choice and the generalized continuum hypothesis.[18]

Several reasons have been given for Gödel choosing NBG for his 1940 monograph.[19] Gödel gave a mathematical reason—NBG's global choice produces a stronger consistency theorem: "This stronger form of the axiom [of choice], if consistent with the other axioms, implies, of course, that a weaker form is also consistent."[20] Robert Solovay conjectured: "My guess is that he wished to avoid a discussion of the technicalities involved in developing the rudiments of model theory within axiomatic set theory."[21] Charles Parsons stated: "This view [that 'property of set' is a primitive of set theory] may be reflected in Gödel's choice of a theory with class variables as the framework for … [his monograph]."[22]

Gödel's achievement together with the details of his presentation led to the prominence that NBG would enjoy for the next two decades.[23] Even Paul Cohen's 1963 independence proofs for ZF used tools that Gödel developed for his work in NBG.[24] However, in the 1960s, ZFC became more popular than NBG. This was caused by several factors, including the extra work required to handle forcing in NBG,[25] Cohen's 1966 presentation of forcing (which uses techniques that naturally belong to ZF),[26] and the proof that NBG is a conservative extension of ZFC.[27]

## Axiomatizating NBG

NBG is presented here as a two-sorted theory, with lower case letters denoting variables ranging over sets, and upper case letters denoting variables ranging over classes. Hence "${\displaystyle x\in y}$" should be read "set x is a member of set y," and "${\displaystyle x\in Y}$" as "set x is a member of class Y." Statements of equality may take the form "${\displaystyle x=y}$" or "${\displaystyle X=Y}$". "${\displaystyle a=A}$" stands for "${\displaystyle \forall x(x\in a\leftrightarrow x\in A)}$" and is an abuse of notation. NBG can also be presented as a one-sorted theory of classes, with sets being those classes that are members of at least one other class.

We first axiomatize NBG using the Template:Visible anchor. This schema is provably equivalent[28] to 9 of its finite instances, stated in the following section. Hence these 9 finite axioms can replace Class Comprehension. This is the precise sense in which NBG can be finitely axiomatized.

### With Class Comprehension schema

The following five axioms are identical to their ZFC counterparts:

pairing implies that for any set x, the set {x} (the singleton set) exists. Also, given any two sets x and y and the usual set-theoretic definition of the ordered pair, the ordered pair (x,y) exists and is a set. By Class Comprehension, all relations on sets are classes. Moreover, certain kinds of class relations are one or more of functions, injections, and bijections from one class to another. pairing is an axiom in Zermelo set theory and a theorem in ZFC.
infinity can be formulated so as to imply the existence of the empty set.[29]

The remaining axioms have capitalized names because they are primarily concerned with classes rather than sets. The next two axioms differ from their ZFC counterparts only in that their quantified variables range over classes, not sets:

The last two axioms are peculiar to NBG:

• Limitation of Size: For any class C, a set x such that x=C exists if and only if there is no bijection between C and the class V of all sets.
From this axiom, due to Von Neumann, Subsets, Replacement, and Global Choice can all be derived. This axiom implies the axiom of global choice because the class of ordinals is not a set; hence there exists a bijection between the ordinals and the universe. If Limitation of Size were weakened to "If the domain of a class function is a set, then the range of that function is likewise a set," then no form of the axiom of choice is an NBG theorem. In this case, any of the usual local forms of Choice may be taken as an added axiom, if desired.
Limitation of Size cannot be found in Mendelson (1997) NBG. In its place, we find the usual axiom of choice for sets, and the following form of the axiom schema of replacement: if the class F is a function whose domain is a set, the range of F is also a set .[30]
This axiom asserts that invoking the principle of unrestricted comprehension of naive set theory yields a class rather than a set, thereby banishing the paradoxes of set theory.
Class Comprehension is the only axiom schema of NBG. In the next section, we show how this schema can be replaced by a number of its own instances. Hence NBG can be finitely axiomatized. If the quantified variables in φ(x) range over classes instead of sets, the result is Morse–Kelley set theory, a proper extension of ZFC which cannot be finitely axiomatized.

### Replacing Class Comprehension with finite instances thereof

An appealing but somewhat mysterious feature of NBG is that its axiom schema of Class Comprehension is equivalent to the conjunction of a finite number of its instances. The axioms of this section may replace the Axiom Schema of Class Comprehension in the preceding section. The finite axiomatization presented below does not necessarily resemble exactly any NBG axiomatization in print.

We develop our axiomatization by considering the structure of formulas.

• Sets: For any set x, there is a class X such that x=X.

This axiom, in combination with the set existence axioms from the previous axiomatization, assures the existence of classes from the outset, and enables formulas with class parameters.

We now turn to quantification. In order to handle multiple variables, we need the ability to represent relations. Define the ordered pair ${\displaystyle (a,b)}$ as ${\displaystyle \{\{a\},\{a,b\}\},}$ as usual. Note that three applications of pairing to a and b assure that (a,b) is indeed a set.

${\displaystyle {\mathit {Conv}}1(R)=\{(b,a)\mid (a,b)\in R\}}$ and
${\displaystyle {\mathit {Conv}}2(R)=\{(b,(a,c))\mid (a,(b,c))\in R\}}$ exist.
• Association: For any class R, the classes:
${\displaystyle {\mathit {Assoc}}1(R)=\{((a,b),c)\mid (a,(b,c))\in R\}}$ and
${\displaystyle {\mathit {Assoc}}2(R)=\{(d,(a,(b,c)))\mid (d,((a,b),c))\in R\}}$ exist.

These axioms license adding dummy arguments, and rearranging the order of arguments, in relations of any arity. The peculiar form of Association is designed exactly to make it possible to bring any term in a list of arguments to the front (with the help of Converses). We represent the argument list ${\displaystyle (x_{1},x_{2},\ldots ,x_{n})}$ as ${\displaystyle (x_{1},(x_{2},\ldots ,x_{n}))}$ (it is a pair with the first argument as its first projection and the "tail" of the argument list as the second projection). The idea is to apply Assoc1 until the argument to be brought to the front is second, then apply Conv1 or Conv2 as appropriate to bring the second argument to the front, then apply Assoc2 until the effects of the original applications of Assoc1 (which are now behind the moved argument) are corrected.

If ${\displaystyle \{(x,y)\mid \phi (x,y)\}}$ is a class considered as a relation, then its range, ${\displaystyle \{y\mid \exists x[\phi (x,y)]\},}$ is a class. This gives us the existential quantifier. The universal quantifier can be defined in terms of the existential quantifier and negation.

The above axioms can reorder the arguments of any relation so as to bring any desired argument to the front of the argument list, where it can be quantified.

Finally, each atomic formula implies the existence of a corresponding class relation:

Diagonal, together with addition of dummy arguments and rearrangement of arguments, can build a relation asserting the equality of any two of its arguments; thus repeated variables can be handled.

#### Mendelson's variant

MendelsonTemplate:Sfn refers to his axioms B1-B7 of class comprehension as "axioms of class existence." Four of these identical to axioms already stated above: B1 is Membership; B2, Intersection; B3, Complement; B5, Product. B4 is Ranges modified to assert the existence of the domain of R (by existentially quantifying y instead of x). The last two axioms are:

B6: ${\displaystyle \forall X\exists Y\forall uvw[(u,v,w)\in Y\leftrightarrow (v,w,u)\in X],}$
B7: ${\displaystyle \forall X\exists Y\forall uvw[(u,v,w)\in Y\leftrightarrow (u,w,v)\in X].}$

B6 and B7 enable what Converses and Association enable: given any class X of ordered triples, there exists another class Y whose members are the members of X each reordered in the same way.

## Discussion

For a discussion of some ontological and other philosophical issues posed by NBG, especially when contrasted with ZFC and MK, see Appendix C of Potter (2004).

Even though NBG is a conservative extension of ZFC, a theorem may have a shorter and more elegant proof in NBG than in ZFC (or vice versa). For a survey of known results of this nature, see Pudlak (1998).

### Model theory

ZFC, NBG, and MK have models describable in terms of V, the standard model of ZFC and the von Neumann universe. Now let the members of V include the inaccessible cardinal κ. Also let Def(X) denote the Δ0 definable subsets of X (see constructible universe). Then:

• Vκ is an intended model of ZFC;
• Def(Vκ) is an intended model of NBG;
• Vκ+1 is an intended model of MK.

### Category theory

The ontology of NBG provides scaffolding for speaking about "large objects" without risking paradox. In some developments of category theory, for instance, a "large category" is defined as one whose objects make up a proper class, with the same being true of its morphisms. A "small category", on the other hand, is one whose objects and morphisms are members of some set. We can thus easily speak of the "category of all sets" or "category of all small categories" without risking paradox. Those categories are large, of course. There is no "category of all categories" since it would have to contain large categories which no category can do. Although yet another ontological extension can enable one to talk formally about such a "category" (see for example the "quasicategory of all categories" of Adámek et al. (1990), whose objects and morphisms form a "proper conglomerate").

On whether an ontology including classes as well as sets is adequate for category theory, see Muller (2001).

## Notes

1. Template:Harvnb. In 1917, Mirimanoff published a form of replacement based on cardinal equivalence (Template:Harvnb).
2. Template:Harvnb. Von Neumann's definition also used the theory of well-ordered sets. Later, his definition was simplified to the current one: An ordinal α is a set that is well-ordered by ∈ and has the property that every member of α is a subset of α (Template:Harvnb).
3. Template:Harvnb (footnote); English translation: p. 398 (footnote).
4. After introducing the cumulative hierarchy, von Neumann could show that Zermelo's axioms do not prove the existence of ordinals α ≥ ω + ω, which include uncountably many hereditarily countable sets. This follows from Skolem's result that Vω+ω satisfies Zermelo's axioms (Template:Harvnb) and from α ∈ Vβ implying α < β (Template:Harvnb); Kunen uses the notation R(β) instead of Vβ.
5. Template:Harfnb. Von Neumann stated his axiom in an equivalent functional form (Template:Harvnb); English translation: p. 400).
6. Fraenkel, Historical Introduction in Template:Harvnb.
7. Template:Harvnb; English translation: p. 399–401.
8. Mirimanoff defined well-founded sets in 1917 (Template:Harvnb).
9. Von Neumann also analyzed Fraenkel's axiom and stated that it was not "precisely formulated", but it would approximately say: "Besides the sets … whose existence is absolutely required by the axioms, there are no further sets." His analysis led him to reject this axiom and to propose the axiom of regularity instead (Template:Harvnb; English translation: p. 404–405 & p. 410–411).
10. Template:Harvnb. In 1930, Zermelo became the first to include regularity in an axiom system (Template:Harvnb).
11. Von Neumann's choice axiom is: "Every relation C has a subclass which is a function and has the same domain." Template:Harvnb.
12. His classes also used "some of the set-theoretic concepts of the Schröder logic and of Principia Mathematica" (quotation from Bernays in Template:Harvnb).
13. Template:Harvnb. Bernays' articles are reprinted in Template:Harvnb.
14. Gödel used von Neumann's axioms in his 1938 announcement of his theorems, and used ZF in his 1939 outline of his proofs (Template:Harvnb).
15. Template:Harvnb. Gödel's consistency proof builds the constructible universe. To build this in ZF requires some model theory; Gödel built it in NBG without model theory. For a discussion of Gödel's technique: see Template:Harvnb.
16. Template:Harvnb, footnote i.
17. Template:Harvnb: "Forcing itself went a considerable distance in downgrading any formal theory of classes because of the added encumbrance of having to specify the classes of generic extensions."
18. Template:Harvnb, p. 107–147. Cohen also gave a detailed proof of Gödel's relative consistency theorems using ZF (Template:Harvnb, p. 85–99).
19. Template:Harvnb, Prop. 4.4, proves Class Comprehension equivalent to the axioms B1-B7 shown on p. 230 and described below.
20. Template:Harvnb, Ex. 4.22(b).
21. Template:Harvnb, axiom R.

## References

• {{#invoke:citation/CS1|citation

|CitationClass=citation }}

• {{#invoke:citation/CS1|citation

|CitationClass=citation }}

• {{#invoke:citation/CS1|citation

|CitationClass=citation }}.

• {{#invoke:citation/CS1|citation

|CitationClass=citation }}.

• {{#invoke:citation/CS1|citation

|CitationClass=citation }}.

• {{#invoke:citation/CS1|citation

|CitationClass=citation }}.

• {{#invoke:citation/CS1|citation

|CitationClass=citation }}.

• {{#invoke:citation/CS1|citation

|CitationClass=citation }}.

• {{#invoke:citation/CS1|citation

|CitationClass=citation }}. (Page numbering in Notes refers to online article whose numbering starts at 1.)

• {{#invoke:citation/CS1|citation

|CitationClass=citation }}.

• {{#invoke:citation/CS1|citation

|CitationClass=citation }}.

• Mendelson, Elliott, (1997), An Introduction to Mathematical Logic, 4th ed. London: Chapman & Hall. ISBN 0-412-80830-7. Pp. 225–86 contain the classic textbook treatment of NBG, showing how it does what we expect of set theory, by grounding relations, order theory, ordinal numbers, transfinite numbers, etc.
• {{#invoke:citation/CS1|citation

|CitationClass=citation }}.

• Richard Montague, (1961), "Semantic Closure and Non-Finite Axiomatizability I," in Infinitistic Methods: Proceedings of the Symposium on Foundations of Mathematics, (Warsaw, 2–9 September 1959). Pergamon: 45-69.
• Muller, F. A., (2001), "Sets, classes, and categories," British Journal of the Philosophy of Science 52: 539-73.
• {{#invoke:citation/CS1|citation

|CitationClass=citation }}.

• Potter, Michael, (2004), Set Theory and Its Philosophy. Oxford Univ. Press.
• Pudlak, P., (1998), "The lengths of proofs" in Buss, S., ed., Handbook of Proof Theory. North-Holland: 547-637.
• {{#invoke:citation/CS1|citation

|CitationClass=citation }}. English translation: {{#invoke:citation/CS1|citation |CitationClass=citation }}.

• {{#invoke:citation/CS1|citation

|CitationClass=citation }}. English translation: {{#invoke:citation/CS1|citation |CitationClass=citation }}.

• {{#invoke:citation/CS1|citation

|CitationClass=citation }}.

• {{#invoke:citation/CS1|citation

|CitationClass=citation }}.