Logique & Analyse 213 (2011), xx

CANTOR-VONNEUMANN SET-THEORY
F.A. MULLER
Abstract
In this elementary paper we establish a few novel results in set the-
ory; their interest is wholly foundational-philosophical in motiva-
tion. We show that in Cantor-Von Neumann Set-Theory, which is
a reformulation of Von Neumanns original theory of functions and
things that does not introduce classes (let alone proper classes),
developed in the 1920ies, both the Pairing Axiom and half the
Axiom of Limitation are redundant the last result is novel. Fur-
ther we show, in contrast to how things are usually done, that some
theorems, notably the Pairing Axiom, can be proved without invok-
ing the Replacement Schema (F) and the Power-Set Axiom. Also
the Axiom of Choice is redundant in CVN, because it a theorem
of CVN. The philosophical interest of Cantor-Von Neumann Set-
Theory, which is very succinctly indicated, lies in the fact that it is
far better suited than Zermelo-Fraenkel Set-Theory as an axioma-
tisation of what Hilbert famously called Cantors Paradise. From
Cantor one needs to jump to Von Neumann, over the heads of Zer-
melo and Fraenkel, and then reformulate.
0. Introduction
In 1928, Von Neumann published his grand axiomatisation of Cantorian Set-
Theory [1925; 1928]. Although Von Neumanns motivation was thoroughly
Cantorian, he did not take the concept of a set and the membership-relation
as primitive notions, but the concepts of a thing and a function for rea-
sons we do not go into here. This, and Von Neumanns cumbersome nota-
tion and terminology (II-things, II.I-things) are the main reasons why ini-
tially his theory remained comparatively obscure. Then came Paul Bernays
[19371953; 1957]. He dressed up Von Neumanns theory in logicist haute
couture, notably with classes (extensions of predicates), and cut out its Can-
torian heart, the Axiom of Limitation (see below). And then, in 1938, came
Gdel. He took this theory of sets and classes as the framework for proving
2 F.A. MULLER
his famous consistency results of the Axiom of Choice and the Generalised
Continuum Hypothesis. Gdel also added the notion of a proper class
as if extensions of predicates (i.e. classes) suddenly stop being extensions
and become improper when they happen to be sets too. The resulting the-
ory of sets and classes, which is usually called Von Neumann-Bernays,
Von Neumann-Bernays-Gdel or even Gdel-Bernays Set-Theory
1
, thus
became known and was used more and more, as time passed, by logicians
and set-theoreticians; it has however remained little known among working
mathematicians other than set-theoreticians or logicians. The standard ax-
iomatisation still is Zermelo-Fraenkel Set-Theory (ZFC), glossing over pos-
sible qualms concerning the Axiom of Choice.
2
Elsewhere we have argued that not ZFC, but what we propose to call
Cantor-Von Neumann Set-Theory (CVN) is the best available axiomatisation
of what Hilbert famously baptised Cantors Paradise.
3
The theory CVN re-
sults when Von Neumanns theory of functions and things is reformulated in
the standard, 1st-order language of pure set-theory (denoted by L

) extended
with a single primitive set V, and certain redundant axioms are deleted.
The purpose of the present note is to write down Cantor-Von Neumann
Set-Theory formally (Section 2), to prove that one-and-a-half axiom is re-
dundant, and to prove some axioms of ZFC in CVN in a manner that differs
from the usual deductions (Section 3). The fact that half of the axiom of
Limitation is redundant has gone unnoticed for about eighty years as far
as this author is aware of. But rst, in order to have some idea what the con-
ceptual watershed between ZFC and CVN consists in, and a fortiori to have a
solid motivation for considering the theory CVN at all, we begin by provid-
ing a very succinct overview of this watershed (Section 1). We emphasise
that the subject of the present paper is not this conceptual watershed, but
a few rigourous results that are the spin-off of a philosophical-foundational
inquiry into Cantorian Set-Theory (cf. Muller [2010]).
1
See Stoll [1963: 318], Fraenkel et al. [1973: 128], Jech [1978: 76], Mostowski in
Mller [1976: 325], Enderton [1977: 10] and Kunen [1980: 35]. To add to the confu-
sion, Fraenkel et al. [1973: 137] call what is almost our Cantor-Von Neumann Set-Theory
G(), where () stands for the Axiom of Limitation (the language L

is then extended
with class-variables to language L

). But this is Von Neumanns theory, not Gdels!

2
Cf. Fraenkel et al.s overview [1973], Ch. II. Ironically, the name Zermelo-Fraenkel
is due to Von Neumann [1961: 321, 348], who also provided the (correct formulation of the)
Axiom of Replacement, who added the Axiom of Regularity, and who created the canonical
theory of ordinal and cardinal numbers; all of this is standardly transplanted to Zermelos
[1908] axiomatisation in order to obtain ZFC.
3
Muller [2010]. This work gratefully builds on Halletts seminal monograph [1984] on
the philosophy and history of Cantorian Set-Theory.
CANTOR-VON NEUMANN SET-THEORY 3
1. Zermelo-Fraenkel versus Cantor-Von Neumann
First comes a string of denitions (which we do not spell out formally
4
): a
set is potential-innite iff it sustains a linear ordering that has no top in the
set; a set is nite iff it can be bijected to 1, 2, 3, . . . , n for some n ,
where is the rst limit Von Neumann-ordinal; a set is actual-innite iff
a proper subset of it can be surjected onto it; a set is ultimate (Quine) iff
no set has it as a member; a set is Cantoresque iff it is not ultimate; set
is absolute-innite iff it is equinumerous to the set of all Cantoresque sets
(Von Neumann); a set is increasable iff it can be surjected onto some more
inclusive set; a set is transnite iff it is actual-innite and increasable. A
set is a Cantor-set (Cantors concept of a Menge) iff it is Cantoresque, in-
creasable, well-founded, well-orderable, has a unique cardinal number, and
has a unique ordinal number as soon as it is ordered well; and, nally, a set
is combinatorially inept iff it does not arise in the cumulative hierarchy.
Standard Zermelo-Fraenkel Set-Theory (ZFC) is dened as the 1st-order
deductive closure of the Axioms of Extensionality, Union, Power, Innity,
Replacement, Regularity and Choice; Replacement entails Separation, and
Replacement and Power together entail Pairing.
5
Cantor-Von Neumann Set-
Theory (CVN) is formulated in L

extended with a single logical constant

V, a primitive set (call this language L
V

); we dene CVN as the 1st-order

deductive closure of the Axioms of Universe (V includes but does not con-
tain every set), Extensionality, Power, Innity, Set-Existence (every predi-
cate with only bounded quantiers, possibly with set-parameters, has a set-
extension of Cantoresque members), Regularity and Weak Limitation (all
ultimate sets are absolute-innite). Von Neumanns Axiom of Limitation is
the conjunction of Weak Limitation and its converse: exactly ultimate sets
are absolute-innite (cf. Section 2 for details).
One proves that in ZFC all sets are Cantor-sets and that neither absolute-
innite nor ultimate sets exist. In CVN one proves that a set is potential-
innite iff it is actual-innite; that a set is ultimate iff it is absolute-innite
(Limitation); that a set is not ultimate iff it is not Cantoresque; that every
set is either absolute-innite or transnite or nite; that every set is either
ultimate or a Cantor-set; and that every set is either increasable or ultimate.
A.H. Kruse essentially proved that CVN is a conservative, hence an equi-
consistent deductive extension of ZFC: every theorem of CVN in which only
4
See Muller [2010] for the formal denitions and the easy proofs of the theorems we are
about to report.
5
See Fraenkel et al. [1973: 22, 52], Lvy [1979: 2324], Suppes [1960: 237],
Stoll [1963: 304].
4 F.A. MULLER
CVN-sets

Cantor-sets
(ZFC-sets)
H
H
H
H
H
H
H ultimate sets
(absolute-innite sets)

transnite sets
H
H
H
H
H
nite sets
Cantor-sets occur already is a theoremin ZFC.
6
Thus in CVN one remarkably
demonstrates, rather than postulates (as in ZFC), that every Cantor-set has
a union-set and a choice-set from axioms that do not smack of choice.
7
Further, CVN is nitely axiomatisable, in contradistinction to ZFC. Model-
theoretically, the whole in every model of CVN that contains exactly the
Cantor-sets is a model of ZFC; and every model of ZFC can be extended to
become a model of CVN, such that the whole of all sets (in the model of
ZFC) becomes the model of V of CVN in the extended model of CVN.
Now, various assertions of Cantor are proved in CVN, whereas they are dis-
proved in ZFC, e.g. that the actual-innite has to be subdivided into the in-
creasable actual-innite and the unincreasable actual-innite
8
(which Can-
tor called the transnite and the absolute-innite, respectively
9
); that
the whole of all Cantor-sets is a perfectly well-dened absolute-innite
set
10
; that every potential innity presupposes an actual-innity
11
; that
an absolute-innite set is mathematically indeterminable
12
(when inter-
preted as combinatorial ineptitude, admittedly with a dosis of wisdom with
6
Cf. Fraenkel et al. [1973: 136137]. Kruse proved this for VN

(see Table); it then

follows that it also holds for CVN.
7
See Fraenkel et al. [1973: 137].
8
Cantor [1932: 375].
9
Cantor [1932: 405].
10
Cantor [1932: 448].
11
Cantor [1932: 410411].
12
Cantor [1932: 375].
CANTOR-VON NEUMANN SET-THEORY 5
hindsight) and cannot be conceived of as a member of another set
13
(they
are ultimate); that absolute-innite sets have to be admitted and acknowl-
edged
14
(they exist); and more. All in all, CVN provides a rigourous legal-
isation for a host of informal claims of Cantor, whereas ZFC outlaws them.
For this reason, CVN is far better suited to be the axiomatisation of Cantors
paradise than ZFC. In a single sentence: to obtain a rigorous legislation of
Cantors Paradise, jump straight to Von Neumann, over the heads of Zermelo
and Fraenkel, and then reformulate.
So much for a brief comparison between ZFC and CVN. We refer to
Muller [2010] for an elaborate inquiry into CVN, its heuristics and moti-
vation, and into how it compares with ZFC conceptually.
2. Axiomatics
To rehearse, the 1st-order formal language L

of pure Set-Theory has only

set-variables (A, B, C, D, F, . . . X, Y, Z; occasionally we use m, n, p as
nite-ordinal-variables), and the membership-relation () as its only dyadic
predicate-constant. The background logic, which is classical 1st-order pred-
icate logic with identity (=). Then

X Y

and

X = Y

are the only types

of atomic sentence. In the language of CVN, denoted by L
V

, we have in ad-
dition to L

one primitive set, V. We use for term-denition, and

iff

for
sentence- and predicate-denition.
Throughout we assume that all the usual denitions are in force (power-set

X of set X, union-set

X of set X, the empty set , etc.; see Fraenkel et
al. [1973], Chapter II). We emphasise that L
V

does not contain Bernaysian

classes, Quinean virtual sets or Gdelian proper classes. We use bold-
faced capitals for ultimate set-names (Kunen-convention). We next spell out
the axioms of CVN formally.
The Universe Axiom says that V includes single set:
(Univ) X : X V . (1)
We introduce the concepts of ultimacy and Cantoresqueness formally, which
are each others negations:
Ultim(X) iff Y : X , Y ; Cantq(X) iff Y : X Y .
(2)
13
Grattan-Guinness [1971: 119].
14
Cantor [1932: 205].
6 F.A. MULLER
Lemma 1: (Ultimacy Lemma) A set is ultimate iff V does not contain it,
or in other words: V is the set of all and only Cantoresque sets: X :
Ultim(X) X V.
Proof (Univ). Let X be ultimate. Then by denition (2), X is not the mem-
ber of every set, V included. Hence X , V.
Let X , V. By Univ (1), Y V for every Y , which means that Z
Y Z V for every Z. Contraposing yields that Z , V Z , Y for
every Z, hence also for X, which permits us to deduce that X , Y . Since
Y is arbitrary, X is ultimate.
The Axiom of Set-Existence (SetEx) asserts that for any predicate (, Y ),
which may have any number of arbitrary set-parameters Y
1
, Y
2
, . . . , Y
n
(ab-
breviated by Y ), and which does not contain unbounded quantiers but may
contain quantiers running over all Cantoresque sets (collected in V), there
exists a set S of all Cantoresque sets for which the predicate holds:
15
(SetEx) Y, S V, X V

X S (X, Y )

. (3)
Notice that in general the set-extension S of SetEx (3) can be ultimate, but
all its members, the sets that fall under the predicate, are Cantoresque (they
cannot be ultimate because they are members, of S); whether set S actually
is ultimate or not is something we have to prove on the basis of the other
axioms. Further, the fact that the variable X in (3) is bounded to V makes it
possible to reduce this list of denumerably many axioms to a eight axioms,
which means that CVNis nitely axiomatisable.
16
The restriction to bounded
quantiers in Set-Existence betrays that it has a whiff of predicativity in
it but certainly not more than a whiff, because quantication over all
Cantorian sets in V still is light-years removed from Russellian typication
and Quinean stratication.
For convenience, we dene V

as the set of all Cantoresque, non-empty

sets (Set-Existence):
V

X V [ Y V : Y X . (4)
That some set exists is a theorem of logic; this set may be V and V may be
empty; to prove that V ,=, another axiom besides SetEx is needed, such as
15
Von Neumann did not proceed from SetEx ab ovo; but remember we announced to erect
a reformulation of Von Neumanns original theory of things and functions.
16
Fraenkel et al. [1973: 129130] for how this is achieved; we are not going to repeat it
hic et nunc.
CANTOR-VON NEUMANN SET-THEORY 7
Innity (which asserts the existence of ); the existence of as a Cantoresque
set follows from Innity, because . Next come the familiar axioms of
Extensionality, Pairing, Union, Power, Innity, Separation and Regularity
(there is implicit universal quantication over all variables that occur free).
(Ext) (X Y Y X) X = Y .
(Pair) X, Y V X, Y V .
(Union) X V

X V .
(Power) X V

X V .
(Inf) V .
(Sep) Z V, Y V, A V,
X

X A X Z (X, Y )

.
(Reg) X V, Y V(Y X Y X =) .
(5)
SetEx (3) provides us already with pair-sets, union-sets and power-sets, and
even with , but nothing can be said as to whether these sets are ultimate
or not; the Axioms of Pairing, Union, Power and Innity decide this by
asserting that these sets are not ultimate.
We employ the usual denitions of a function F from its domain D to
co-domain C, denoted by F : D C:
F : D C iff X D, Y C : X, Y ) F . (6)
So F D C. The range of F is the set of everything reached by F from
D:
R
F
Y C [ X D : X, Y ) F . (7)
Then R
F
C. The Axiom of Replacement then reads that for every func-
tion F from domain D to co-domain C it holds that if its domain is Can-
toresque, then so is its range:
(F) (F : D C) (D V R
F
V) . (8)
The Axiom of Global Choice reads that there is some function F V
(also called a choice-function) that sends every non-empty set to a member
of it:
(GChoice) F V

F : V

X V

: F(X) X

. (9)
8 F.A. MULLER
GChoice (9) implies Choice as we know it from ZFC: restrict F in (9) to
Cantoresque subsets of V

(replace

F V

with

F V

).
We now arrive at the Cantorian heart of CVN. Von Neumann essentially
proposed two precise renditions of Cantors idea of an absolute-innite
set: rst, as a set that cannot be collected further into any other set, which
is of course ultimacy (2); and secondly, as being absolute-innite, dened as
being equinumerous to the set V of all Cantoresque sets:
AbsInf(X) iff X V , (10)
where is the equinumerosity-relation. Denition: set X is equinumerous
to set Y iff there is bijection between them:
X Y iff F V : X Y , (11)
where a bijection F from X to Y , denoted by F : X Y , is dened as a
function whose range Y is such that every member of Y comes from exactly
one domain-member:
F : X Y
iff

F : X Y B Y, A X : A, B) F

.
(12)
Thus for Von Neumann, too big means most encompassing: the only way
for a set to become absolute-innitely big is to be as big as what encompasses
everything. Since Von Neumann intended ultimacy as a new way of looking
at absolute-innity, whence the following axiom
Axiom 2: (Limitation) All and only the ultimate sets are absolute-innite:
X: Ultim(X) AbsInf(X).
The Weak Axiomof Limitation asserts one conjunct of Limitation (Axiom2).
Axiom 3: (Weak Limitation: Absolute-Ininity of the Ultimate) Every ulti-
mate set is absolute-innite: X : Ultim(X) AbsInf(X).
For the sake of reference and overview, we dene the following theories,
as the 1st-order deductive closures of the axioms in the language mentioned:
CANTOR-VON NEUMANN SET-THEORY 9
Theory Language Axioms Theorems
ZFC L

Ext, Inf, Reg, Pow, Un, F, Choice Sep, Pair

CVN
0
L
V

SetEx, Ext, Inf, Reg, Univ

CVN L
V

SetEx, Ext, Inf, Reg, Univ, Pow, GChoice, F, Sep,

WkLim Un, Pair, Lim
VN

SetEx, Ext, Inf, Reg, Pow, Un, GChoice, F, Sep

Pair, Lim (Un, Pair)
VN

comes closest to Von Neumanns original theory (VN), when reformu-

lated in L

enriched with class-variables in Bernaysian fashion (L

)
17
; Un
and Pair between brackets in the Theorems-column indicate they need not
be taken as axioms, as Von Neumann originally did, because they can be
proved on the basis of the other axioms. For reasons indicated in Section 1,
we baptise the theory in the third row of the Table above Cantor-Von Neu-
mann Set-Theory (CVN).
Axiom 2 of Limitation provokes the question whether it is not some philo-
sophical ornament, solely put forward by Von Neumann to propitiate the
Cantorian spirit. The answer is a resounding denial, for Limitation is, in
the presence of CVN
0
plus Power and Pairing, equivalent to the conjunction
of Global Choice, Separation, Replacement (this was essentially proved by
Von Neumann [1928]) and Union (proved by Lvy [1968])! This is one ex-
cellent reason why Von Neumann adopted the theory VN (with Union), be-
cause then Global Choice, Separation and Replacement become theorems;
consequently one then nds ZFC among its deductive offspring so that VN
deductively extends ZFC. Thus calling the Axiom of Limitation, perhaps pe-
joratively, a Cantorian ornament does not even begin to do justice to it.
Besides its deductive strength, Von Neumann motivated the Axiom of Limi-
tation on two independent grounds: (i) it captures Cantors notion of an un-
increasable, actual-innite set and recognises and admits their existence
(all Cantors words); and (ii) it blocks the deduction of the well-known antin-
omies (Russell, Burali-Forti) and simultaneously, seemingly per impossibile,
it almost saves the Peano-Frege principle of full comprehension according
to which every predicate has an extension (by binding variables mildly to
V). If all this is ornamental, then we better reconstrue mathematics as the
17
Fraenkel et al. [1973: 128] call VN

without Limitation Von Neumann-Bernays

(VNB) and VN

with Choice VNBC. Of course they employ a two-sorted language, and

moreover add redundant axioms: Union, Pairing, Replacement, Separation
10 F.A. MULLER
rigorous inquiry into ornaments.
From the axioms we turn to the theorems.
3. Weak Limitation and its Consequences
By means of a series of theorems, we shall work our way to the central
result: CVN entails VN, which is to say that Pairing and the converse of
Weak Limitation can be proved in CVN:
CVN WkLim, Union . (13)
Of course, as soon as we have Limitation, we have Union, Separation, Re-
placement and (Global) Choice, due to Von Neumanns result mentioned
earlier, but it is interesting to know, we believe, that several of these theo-
rems can be proved without invoking Limitation. We commence with
Lemma 4: Set V contains every set that is included or contained in some
Cantoresque set:
Y V(X Y X Y ) X V .
Proof. It is sufcient to prove that both disjuncts separately imply that
X V. (i) If X Y V, then X is contained in some set and there-
fore not ultimate (2), which is the same as being a member of V according
to Lemma 1. (Thus V is a transitive set.)
(ii) If X Y V, then X

Y V (Power) and then the transitivity of
V (i) yields X V.
Theorem 5: (Ultimacy of V) Set V is ultimate.
Proof (CVN
0
, Power; Innity of CVN
0
will not be employed.) We prove it
by reductio ad absurdum. Assume that V is Cantoresque (Reductio Assump-
tion; henceforth: RA).
To steer at a contradiction, we remark that RA entails that V is self-mem-
bered because V contains by Lemma 1 all Cantoresque sets:
(i) V V .
Pow guarantees that the singleton-set X of every Cantoresque set X
V (whose existence follows from SetEx) is Cantoresque, because X

X V. Then by (RA) V V, because of Lemma 1. We obviously

CANTOR-VON NEUMANN SET-THEORY 11
also have by denition of the singleton-set of V:
(ii) V V .
Regularity, when applied to Cantoresque set V, now says there is some
Y V such that V and Y have no members in common. This Y must
be V, because V is the only member of V. So V and V have no members
in common. But according to (i) and (ii), they do have a member in common:
V. Contradiction.
In ZFC, Pairing (5) is proved on the basis of Power and Replacement; we
prove it here without using any of these axioms. We prove it on the basis of
CVN
0
enriched with WkLim (see Table).
Theorem 6: (Paring) For every pair of Cantoresque sets, their pair-set exists
and is Cantoresque.
Proof (CVN
0
, WkLim). Let X, Y V. By Set-Existence and Extension-
ality, the pair-set X, Y V (Univ) exists uniquely. To prove that it is a
member of V, assume, for reductio, that X, Y is not a member of V, i.e.
that it is ultimate (RA): X, Y , V.
Then X, Y is equinumerous to V, due to Weak Limitation (Axiom-3).
Let f be the bijection from V onto X, Y , the existence of which is by
denition (11) logically equivalent to the equinumerosity of X, Y and V.
Well, Innity gives us V (Univ). Then V. Let m, n be distinct
(n ,= m). If f(m) or f(n) is not equal to X or Y , then f is not even a
function with range X, Y and we already have a contradiction. To avoid
this contradiction, we must have that f(m) = X and f(n) = Y , or with X
and Y interchanged. Obviously, there is a p that is distinct from both n
and m. Hence there must be a member of X, Y , say Z, distinct from X
and Y , such that f(p) = Z. But X, Y has no other members besides X
and Y . Contradiction. So if X, Y V, then X, Y V.
Now that we have Pairing in CVN
0
plus WkLim, we can make pairs, or-
dered pairs and sets of ordered pairs, etc. All functions, relations and op-
erations are now available. Notice that neither Replacement nor Power was
needed to prove Pairing, whereas in ZFC precisely these, and only these,
axioms are used to prove Pairing these proofs do not carry over to CVN
because we do not have Replacement (yet) and our proof of Replacement (as
well as Von Neumanns proof) requires the presence of Pairing.
18
Instead of
18
Von Neumann had Pairing as an axiom [1928; 1961: 344]. To prove Pairing from
Replacement and Power, consider , which exists according to Separation (a consequence
12 F.A. MULLER
Inf, Power can be used in the proof above to produce more than 2 members
of V via

A,

B,

A,

B etc. so as to get at a contradiction.

The absolute-innity of V is an immediate consequence of Theorem5 as
soon as we have Weak Limitation. But it can be proved without explicitly
appealing to Weak Limitation, on the basis of Universe, Extensionality, Set-
Existence and Pairing.
Theorem 7: (Absolute-Innity of V) Set V is absolute-innite.
Proof (Univ, Ext, SetEx, Pairing). We prove something stronger: every set is
equinumerous to itself, hence V included. First we dene for an arbitrary set
X V, the identity
X
V as the diagonal set of ordered pairs A, A),
for all A X. The unique existence of
X
is guaranteed by SetEx, Ext and
Pairing. According to denition (11), X is equinumerous to itself iff there
is a bijection between set X and itself, i.e. there is a set of ordered pairs
A, B), where A, B X, such that every A X and every B X occur
exactly once. Clearly the diagonal set
X
qualies as such a set.
The equinumerosity of V with itself yields its absolute-innite character
by denition (10).
As a corollary of Theorems 5 and 7, we have an instance of the Axiom of
Limitation for V (theorem of logic: entails ):
CVN
0
, WkLim Ultim(V) AbsInf(V) . (14)
Now we prove Zermelos Axiom Schema of Separation.
Theorem 8: (Separation Schema) Every predicate has a Cantoresque set-
extension of members in a given Cantoresque set:
CVN
0
, Pow Z V, Y V, A V,
X

X A X Z(X, Y )

.
Proof (CVN
0
, Pow; of CVN
0
neither Regularity nor Innity will be used).
Let Z be an arbitrary Cantoresque set, Z V, and (, Y ) some sentence
with one free variable and n set-parameters, abbreviated by Y . Set-Existence
gives us the set-extension A of predicate

(X, Y )X Z

, which has Z
of Replacement); make = {{}, }, which is a Cantoresque set (Pow); biject it to the set
{A, B}, where Aand B are distinct but further arbitrary sets; and nally invoke Replacement
to conclude that {A, B} also is Cantoresque.
CANTOR-VON NEUMANN SET-THEORY 13
as an additional Cantoresque set-parameter. Is A a Cantoresque set? Yes, by
application of Lemma 4.
The Schema of Separation can be replaced with a single sentence (notably
this cannot be done in ZFC, where V is not available): a subset of a Can-
toresque set is Cantoresque; formally,
X V (X V) V . (15)
Since it is a theorem of logic that (X V) X, Lemma 4 does the rest.
Denitions: X is minumerous or equinumerous to Y , denoted by X Y ,
iff X can be bijected to a subset of Y ; and X is minumerous to Y , or synony-
mously, Y is amplinumerous to X, denoted by X Y , iff X is minumerous
or equinumerous to Y and Y is not minumerous or equinumerous to X:
X Y iff Z Y, F V : X Z .
X Y iff X Y

(Y X) .
(16)
We report two theorems: Cantors Power Theorem, according to which ev-
ery set is minumerous to its power-set, and the Cantor-Dedekind-Bernstein
Minumerosity Theorem, which asserts the a-symmetry of the relation :
CVN
0
, Pow X

X
CVN
0
, Pow (X Y Y X) X Y .
(17)
The following theorem directly follows from the denition of minumer-
osity (16) and Theorem(17):
CVN X Y

(Y X) X , Y

X Y X , Y

.
(18)
Cantors Power Theorem(17) can be proved on the basis of SetEx (Separa-
tion sufces), Ext, Power and Pairing, which together yield that (a) X

X
(easy: A A bijects X onto a subset of

X); and (b) X ,

X (by
means of a reductio argument); in the nal step of the proof, the Minu-
merosity Theorem is invoked, via version (18), to deduce from (a) and (b)
that X

X.
19
The Minumerosity Theorem can be proved from Sep, Ext
19
See the proofs of Cantors Power Theorem and the Minumerosity Theorem in, for in-
stance, Stoll [1963: 8182, 86], Lvy [1979: 85, 87].
14 F.A. MULLER
and Pair, hence also in CVN, which has Sep (Thm 8) and Pairing (Thm 6) as
theorems (in ZFC), and Ext as an axiom.
Now we are in a position to prove the converse of Weak Limitation in
CVN.
Theorem 9: (Ultimacy of the Absolute-Innite) Every absolute-innite set
is ultimate and therefore not Cantoresque.
Proof (CVN). Let X be an absolute-innite set: X V (Assumption). We
have to prove that X is ultimate. The reductio assumption is that X is not
ultimate: X Y for some Y (RA).
When X is not ultimate, then by Lemma 1, X V, and then

X V
(Pow). Every set Y V can be bijected to a subset of V, namely to it-
self by means of the identity function; hence Y V. In combination with
Y

Y (17), we then deduce that X V. By means of (18), we conclude
that V , X, which contradicts the Assumption.
Every axiom of CVN is invoked to prove that absolute-innite sets are
ultimate (Thm9), all via Pairing (Thm 6) and the Minumerosity Thm (17).
We then arrive at:
Theorem 10: (Limitation Theorem) In CVN, the absolute-innite sets are
exactly the ultimate sets.
From the Limitation Theorem (Thm10) and Pairing (Thm6), our main result
follows: CVN entails VN(13).
The Axiom of Replacement, which in ZFC is an axiom schema, is reduced
in CVN to a single sentence of L

:
Theorem 11: (Replacement) If the domain of a function is Cantoresque, then
so is its range (8).
Proof (CVN). We prove it on the basis of CVN; we submit the proof as a
simpler one than Von Neumanns proof [928; 1961: 365]; only WkLim will
be involved.
Let F be a function whose domain D
F
is Cantoresque: D
F
V (Assump-
tion). Dene the set of members of D
F
which F sends to a given member
Y R
F
(SetEx, Ext):
[Y ]
F
X D
F
[ F(X) = Y . (19)
Since [Y ]
F
D
F
, and thus [Y ]
F


D
F
(Pow), we may conclude that
[Y ]
F
is Cantoresque (for every Y R
F
). We next collect them in a set
CANTOR-VON NEUMANN SET-THEORY 15
(SetEx, Ext):
Z
F

[Y ]
F


D
F
[ Y R
F

.
Then Z
F


D
F
, which implies that
(i) Z
F

D
F
.
The range R
F
is equinumerous to Z
F
, because Y [Y ]
F
is bijective
from R
F
to Z
F
. When we combine R
F
Z
F
with (i), we obtain that
R
F

D
F
. From this and

D
F


D
F
(17) and

D
F
V (because

D
F
V), we then have that R
F
V. By virtue of (18), we then have
(ii)

(V R
F
) .
If R
F
were equinumerous to V, then we would trivially have that V R
F
,
in contradiction to (ii); hence V is not equinumerous to R
F
. But then, by
WkLim (3), R
F
is not ultimate either. Hence if D
F
V, then R
F
V.
We nally consider our last theorem.
Theorem 12: Cantor-Von Neumann set-theory has Union and Global Choice
as theorems.
Proof Sketch (CVN). Lvy [1968] surprisingly proved Union, which had been
considered as an unprovable axiom for more than forty years. The proof
crucially employs Replacement and further WkLim; it carries over to CVN.
We sketch it.
Set is the largest ordinal number, which means that is a set well-
ordered by . Since is ultimate (Burali-Forti), it is absolute-innite (Wk-
Lim) and therefore equinumerous to V. Then there is a one-one correspon-
dence F : V, F(). Call its inverse : V , X (X);
let (X) be the least ordinal such that X = F((X)).
First it is proved, using Replacement, that every set of ordinals is bounded
by some ordinal (this is essentially a result from ordinal arithmetic: the
ordinal-sum of an arbitrary number of ordinal numbers exist); then the set
of ordinals is a subset of this bound, because for every , : if
( or = ), then . SetEx yields the existence of set

X
for every Cantoresque set X. The challenge is now to establish that

X is
Cantoresque.
Replacement gives us the Cantoresque set of all Cantoresque sets [Y ] for
sets Y

X. The set of all sup[Y ] , i.e. the least upperbound for the
set of ordinals [Y ], for Y X, is bounded by some Cantoresque ordinal,
say. Then Cantoresque set F[]

X. Replacement yields that set

X,
16 F.A. MULLER
then, is also Cantoresque. So

X V whenever X V. So much for
Union; next Global Choice.
The bijection F : V, considered above, can be exploited to show
that V can be well-ordered.
20
Loosely speaking, since every ordinal
is by denition well-ordered by , so that we have a woset ,

), we can
replace every member , ) of the membership-relation

on , that is, of

, ) [

, (20)
with the pair F(), F()); this replacement yields the set:

F(), F()) V V [ , )

. (21)
The set

is Cantoresque (due to Replacement) and is a well-ordering on

the set F(): F(),

) is a woset. The images of the unique

-bottoms
of every woset F() can be collected in a set equinumerous to V (because
F is bijective), which results in a choice-set of V.
The main conclusions of this paper are, besides that Pairing is redun-
dant axiom in Cantor-Von Neumann Set-Theory (CVN), that, rst, half of
Von Neumanns Axiom of Limitation, which is the Cantorian heart of CVN,
is redundant; and secondly, that the proofs of the important theorems in CVN
(Pairing, Separation, Replacement, Global Choice, Union) reveal that this
half of Limitation (which we call Weak Limitation) is the part of Limita-
tion that performs all the deductive labour. The strength of Weak Limitation
is as Herculean as Von Neumanns intellectual powers were.
Fac. of Philosophy H5-10
Erasmus University Rotterdam
3062 PA The Netherlands
and Institute for the History and Philosophy of Science
Utrecht University
E-mail: [email protected]
REFERENCES
[Muller 2010] F.A., Cantors Paradise and Von Neumanns Theory, L
A
T
E
X-
script of a monograph, work in progress; based on Muller [1998].
[Muller 1998] F.A., Structures for Everyone (disseration for doctorate, Uter-
cht University); Amsterdam: Gerits & Son, 1998.
20
Neumann [1928: 720721].
CANTOR-VON NEUMANN SET-THEORY 17
[Bernays 19371954] P., A System of Axiomatic Set Theory, published in
The Journal of Symbolic Logic over the years 19371954; collected in
Mller [1976], 1119.
[Bernays 1958] P., Axiomatic Set Theory, Amsterdam: North-Holland Pub-
lishing Company, 1958.
[Cantor 1932] G., Gesammelte Abhandlungen mathematischen und
philosophischen Inhalts, E. Zermelo (ed.), Berlin: Springer-Verlag,
1932.
[Enderton 1977] H.B., Elements of Set Theory, New York-London: Aca-
demic Press, 1977.
[Fraenkel 1973] A.A., Y. Bar-Hillel, A. Lvy and D. van Dalen, Foundations of
Set Theory (2nd Ed.), Amsterdam: North-Holland, 1973.
[Grattan-Guinness 1971] I., The Correspondence between Georg Cantor
and Philip Jourdain, Jahresbericht der deutschen Mathematiker Vere-
inigung 73 (1971) 111130.
[Hallett 1984] M., Cantorian Set Theory and Limitation of Size, Oxford:
Clarendon Press, 1984.
[Jech 1978] T., Set Theory, New York: Academic Press, 1978.
[Kunen 1980] K., Set Theory. An Introduction to Independence Proofs, Am-
sterdam: North-Holland, 1980.
[Lvy 1968] A., On Von Neumanns axiom system for Set Theory, Amer-
ican Mathematical Monthly 75 (1968) 763763.
[Lvy 1979] A., Basic Set Theory, Berlin: Springer-Verlag, 1979.
[Mller 1976] G.H., Sets and Classes. On the work of Paul Bernays, Ams-
terdam: North-Holland Publishing Company, 1976.
[Neumann 1925] J. von, Eine Axiomatisierung der Mengenlehre, Journal
fr die Mathematik 154 (1925) 219240; also in Neumann [1961].
[Neumann 1928] J. von, Die Axiomatisierung der Mengenlehre, Mathe-
matische Zeitschrift 27 (1928) 669752; also in Neumann [1961].
[Neumann 1929] J. von, ber eine Widerspruchfreiheitsfrage der Axioma-
tischen Mengenlehre, Journal fr die Mathematik 160 (1929) 227
241; also in Neumann [1961].
[Neumann 1961] J. von, Collected Works I, A.H. Taub (ed.), Oxford: Perg-
amon Press, 1961.
[Stoll 1963] R.R., Set Theory and Logic, USA: W.H. Freeman, 1963.
[Suppes 1960] P., Axiomatic Set Theory, Princeton, New Jersey: Van Nos-
trand, 1960.
[Zermelo 1908] E., Untersuchungen ber die Grundlagen der Mengen-
lehre, Mathematische Annalen 65 (1908) 261281.