# Morse–Kelley set theory

In the foundations of mathematics, **Morse–Kelley set theory** (**MK**), **Kelley–Morse set theory** (**KM**), **Morse–Tarski set theory** (**MT**), **Quine–Morse set theory** (**QM**) or the **system of Quine and Morse** is a first-order axiomatic set theory that is closely related to von Neumann–Bernays–Gödel set theory (NBG). While von Neumann–Bernays–Gödel set theory restricts the bound variables in the schematic formula appearing in the axiom schema of Class Comprehension to range over sets alone, Morse–Kelley set theory allows these bound variables to range over proper classes as well as sets, as first suggested by Quine in 1940 for his system ML.

Morse–Kelley set theory is named after mathematicians John L. Kelley and Anthony Morse and was first set out by Wang (1949) and later in an appendix to Kelley's textbook *General Topology* (1955), a graduate level introduction to topology. Kelley said the system in his book was a variant of the systems due to Thoralf Skolem and Morse. Morse's own version appeared later in his book *A Theory of Sets* (1965).

While von Neumann–Bernays–Gödel set theory is a conservative extension of Zermelo–Fraenkel set theory (ZFC, the canonical set theory) in the sense that a statement in the language of ZFC is provable in NBG if and only if it is provable in ZFC, Morse–Kelley set theory is a proper extension of ZFC. Unlike von Neumann–Bernays–Gödel set theory, where the axiom schema of Class Comprehension can be replaced with finitely many of its instances, Morse–Kelley set theory cannot be finitely axiomatized.

NBG and MK share a common ontology. The universe of discourse consists of classes. Classes that are members of other classes are called sets. A class that is not a set is a proper class. The primitive atomic sentences involve membership or equality.

With the exception of Class Comprehension, the following axioms are the same as those for NBG, inessential details aside. The symbolic versions of the axioms employ the following notational devices:

A set and a class having the same extension are identical. Hence MK is not a two-sorted theory, appearances to the contrary notwithstanding.

**Foundation:** Each nonempty class *A* is disjoint from at least one of its members.

**Limitation of Size:** *C* is a proper class if and only if *V* can be mapped one-to-one into *C*.

The formal version of this axiom resembles the axiom schema of replacement, and embodies the class function *F*. The next section explains how Limitation of Size is stronger than the usual forms of the axiom of choice.

**Power set:** Let *p* be a class whose members are all possible subsets of the set *a*. Then *p* is a set.

Note that *p* and *s* in Power Set and Union are universally, not existentially, quantified, as Class Comprehension suffices to establish the existence of *p* and *s*. Power Set and Union only serve to establish that *p* and *s* cannot be proper classes.

Monk (1980) and Rubin (1967) are set theory texts built around MK; Rubin's ontology includes urelements. These authors and Mendelson (1997: 287) submit that MK does what is expected of a set theory while being less cumbersome than ZFC and NBG.

MK is strictly stronger than ZFC and its conservative extension NBG, the other well-known set theory with proper classes. In fact, NBG—and hence ZFC—can be proved consistent in MK. MK's strength stems from its axiom schema of Class Comprehension being impredicative, meaning that φ(*x*) may contain quantified variables ranging over classes. The quantified variables in NBG's axiom schema of Class Comprehension are restricted to sets; hence Class Comprehension in NBG must be predicative. (Separation with respect to sets is still impredicative in NBG, because the quantifiers in φ(*x*) may range over all sets.) The NBG axiom schema of Class Comprehension can be replaced with finitely many of its instances; this is not possible in MK. MK is consistent relative to ZFC augmented by an axiom asserting the existence of strongly inaccessible cardinals.

The only advantage of the axiom of limitation of size is that it implies the axiom of global choice. Limitation of Size does not appear in Rubin (1967), Monk (1980), or Mendelson (1997). Instead, these authors invoke a usual form of the local axiom of choice, and an "axiom of replacement,"^{[1]} asserting that if the domain of a class function is a set, its range is also a set. Replacement can prove everything that Limitation of Size proves, except prove some form of the axiom of choice.

The class of von Neumann ordinals can be well-ordered. It cannot be a set (under pain of paradox); hence that class is a proper class, and all proper classes have the same size as *V*. Hence *V* too can be well-ordered.

MK can be confused with second-order ZFC, ZFC with second-order logic (representing second-order objects in set rather than predicate language) as its background logic. The language of second-order ZFC is similar to that of MK (although a set and a class having the same extension can no longer be identified), and their syntactical resources for practical proof are almost identical (and are identical if MK includes the strong form of Limitation of Size). But the semantics of second-order ZFC are quite different from those of MK. For example, if MK is consistent then it has a countable first-order model, while second-order ZFC has no countable models.

ZFC, NBG, and MK each have models describable in terms of *V*, the von Neumann universe of sets in ZFC. Let the inaccessible cardinal κ be a member of *V*. Also let Def(*X*) denote the Δ_{0} definable subsets of *X* (see constructible universe). Then:

MK was first set out in Wang (1949) and popularized in an appendix to J. L. Kelley's (1955) *General Topology*, using the axioms given in the next section. The system of Anthony Morse's (1965) *A Theory of Sets* is equivalent to Kelley's, but formulated in an idiosyncratic formal language rather than, as is done here, in standard first-order logic. The first set theory to include impredicative class comprehension was Quine's ML, that built on New Foundations rather than on ZFC.^{[2]} Impredicative class comprehension was also proposed in Mostowski (1951) and Lewis (1991).

The axioms and definitions in this section are, but for a few inessential details, taken from the Appendix to Kelley (1955). The explanatory remarks below are not his. The Appendix states 181 theorems and definitions, and warrants careful reading as an abbreviated exposition of axiomatic set theory by a working mathematician of the first rank. Kelley introduced his axioms gradually, as needed to develop the topics listed after each instance of *Develop* below.

Notations appearing below and now well-known are not defined. Peculiarities of Kelley's notation include:

Identical to *Extensionality* above. **I** would be identical to the axiom of extensionality in ZFC, except that the scope of **I** includes proper classes as well as sets.

'α' and 'β' are replaced by variables, ' *A* ' by a formula Æ, and ' *B* ' by the formula obtained from Æ by replacing each occurrence of the variable that replaced α by the variable that replaced β provided that the variable that replaced β does not appear bound in *A*.

*Develop*: Boolean algebra of sets. Existence of the null class and of the universal class *V*.

The import of **III** is that of *Power Set* above. Sketch of the proof of Power Set from **III**: for any *class* *z* that is a subclass of the set *x*, the class *z* is a member of the set *y* whose existence **III** asserts. Hence *z* is a set.

*Develop*: *V* is not a set. Existence of singletons. Separation provable.

*Develop*: Unordered and ordered pairs, relations, functions, domain, range, function composition.

**V. Substitution:** If *f* is a [class] function and *domain f* is a set, then *range f* is a set.

The import of **V** is that of the axiom schema of replacement in NBG and ZFC.

The import of **VI** is that of *Union* above. **IV** and **VI** may be combined into one axiom.^{[3]}

*Develop*: Cartesian product, injection, surjection, bijection, order theory.

*Develop*: Natural numbers, **N** is a set, Peano axioms, integers, rational numbers, real numbers.

**IX** is very similar to the axiom of global choice derivable from *Limitation of Size* above.

*Develop*: Equivalents of the axiom of choice. As is the case with ZFC, the development of the cardinal numbers requires some form of choice.

If the scope of all quantified variables in the above axioms is restricted to sets, all axioms except **III** and the schema **IV** are ZFC axioms. **IV** is provable in ZFC. Hence the Kelley treatment of **MK** makes very clear that all that distinguishes **MK** from ZFC are variables ranging over proper classes as well as sets, and the Classification schema.