Ackermann's Set Theory

ANNALS Oi: MATHEMATICAL LOGIC -~ Volume 2, No. 2 (1970) pp.
1 8 9 - 2 4 9
ACKERMANN's SET THEORY EQUALS Z F *
William N,REINHARDT
Boulder. Colorado
Received 26 Janu,'~y 1970
Section 0
! ritroduction
Section 0.1 introduces Ackermann's set theory and contains a brief

summary of the history of work on the theory. Section 0.2 contains a
summary of the results of tlais paper. Sections 0.3, 0.4, and 0.5 contain
preliminaries for the later sections (notation, etc.). All notions used in
0.1 and 0.2 will be fully defined in 0.3, 0.4, 0.5.
O. 1. A ckermann's set theor.v

The set theory of Ackermann is formulated in a first order language
wittl one binary relation symbol ~, and one individual constant V. The
objects the variables range over are referred to as classes, 'x E y ' is read
'x is a member of y', and 'x ~ V' is read 'x is a set'. Tiros the theory dis-
tinguishes between sets and classes; every set is a class but some classes
are not sets. (It is not assumed that every member of a class is a set.)
The axioms are the universal closures of
* This paper contains a major part of the author's doctoral dissertation. 1 wish to thank the fol-
lowing people for helpful conversations eoncc'.ning this work: R.L.Vaught, my thesis advisor
(who also brougl~t to my attention most of fl~e y~ ~blems considered), J.Malitz, A.Levy, and
J.Silver. I also xvish to thank the Danforth Foundation for encouragement and support. Part
of this work was supported by NSF grants, including GP 12209, GP 7387.
190 A ct,'ermaml "s set thcorr equcls ZF
A1. (Extensionality)
At(t~x ~ t ~j,)~ x =y.
A2. (Class construction)
Vz A t ( t ~ z * - - ~ t ~ V ^ ® ) ,
where O is any formula in ~ and V not involving z.

A3. (Strong completeness of V)
xE V^ [tEx v Att(uE t ~ uEx)] ~ tE V.
A4. (Ackermann's schema)
x,y EVA At(O(x,y, t)* 1E V
-* V z ~ V A t ( t E z ÷ ~ O ( x , y , t))
where the free variables of 0 are x, ~', t, and V does not occur in O.
We will let A be the theory determined by these axioms. A* is o1: ~ained

from A by adding the universal closure of the axiom.
A5. (Regularity for sets)
x~V^Vu(u~x)
--, Vt~(u ~ x / x A t ( t ~ x -, t q~ , ) ) .
Perhaps some remarks on the motivation for the theory A will be

helpful. Our discussion will center on A4; AI, A2 are familiar and A3
says that any member of a set is a set, and that any subclass of a set is a
set. Ackermann interpreted the 'definite' in Cantor's well known defi-
nition ("By a set we understand any collection of definite distinct ob-
jects ... into a whole" [31, p. 282) as requiring t~:at in order for a col-
lection of sets to be a set, "what belongs to the collection and what
does not belong to it must be sufficiently shaprly delimited". (Acker-
mann [ t ], p. 337.) Ackerl~aann considered that the concept of set
itself" is thoroughly open, however, and thus not sharply delimited. The
basis for A4 appears to be the idea that sufficiently sharply delimited
collections of sets are sets. The hypothesis of A4 assures that the prop-
In m~ductiol~ 191
erty 't is a set such that ,;(,v, y, t)' is in a certain sense independent of
the (extension of t h e ) c o n c e p t of set, and may be regarded as giving a
sufficient condition for a collection to be sharply delimited. (A further
dis,,:ussion of these ideas is contained in Section 5.~
Tile intuitive ideas leading to Ackermann's theory are thus seen to
be quite different from those which lead to Zermelo-Fraenkel se~
theory (ZF). In that theory, of course, there are a number of set con-
struction principles, and there is an appeal to the intuitive principle
that a collection which is equivalent in power to a set must be a set.
Thus it is not obvious how the two theories compare.
It should bc remarked that the construction !~rinciples of ZF were
originally fornmlated to reflect the practice of mathematicians; rather
than starting from first principles and deriving set theory, one started
with set theory (a'- it had been developed historically) and tried to find
principles adequate to develop th,: theory which did not yield the
known paradoxes: no attempt was made to provide a philosophical
explanation of the source of the principles, or a clarification of their
intuitive basis (see E.Zemaelo 12! l P. 26t, 262). For a contemporary
discussion of intuitive ideas which, provide a basis for Zermelo's axiom's,
see Kreisel 122], pp. 100-105.
Much of the work that has been done on Ackermann's theory has
been directed toward comparing A and ZF.
(1) In the paper introducil.g the theory A [1 ], Ackermann showed that

the retativizations to V of all of the axioms of Zermelo's theory (Z)
(except for the regularity axiom) could be proved in A. He also stated
that the replacement schema of ZF (relativized to V) can be proved in
A (compare (2) below). An error in Ackermann's proof of this asser-
tion was found by Levy [81, however, and the question remained open.
Ackermann also introduced an extension Aoo of A which appears much
stronger than A: instead of a class V of all sets, there is for each natural
number n a class V,z of sets of 'type n', together with appropriate
axioms (for a version of Ao~ see Chapter 4).
(11) A.kevy showed (18]) that if there is an inaccessible cardinal, then
there is a natural model of A (that is, one oi the Corm ( RCL e, Ra>,
where ~ </3 are ordinals and R/3 consists ot the sets of rank less than t3).
He also proved that
192 A ckermann's set theory equals ZF
(1) If the relativization to V of a sentence o is provable in A*, then

o is provable in ZF,
thus showing that A* gives us no more information about the sets than
ZF does. These results were also proved (with appropriate modifications)
for Aoo. Levy proved a number of theorems within A, and in particular
showed that a strong reflection principle holds in A*. Thus Levy showed
that (the relativizations to V of) very strong theorems of ZF could be
proved in A*, and he raised the question whether the converse of (1)
holds.
(1II) Levy and Vaught [10l developed the theory of A* further, show-
ing that still larger classes of (the relativizations to V of) tt'eorems of
ZF could be proved in A*.
(IV) In his thesis [6], R.Grewe succeeded in obtaining a partial con-
verse to (1): he showed that in every natural model (R/3, c, Rc~>of A,
(R0~, e) ~ ZF (thus answering Levy's question for the case of natural
models of A*). This result is included in the paper [6a].
Grewe also introduced a methof of interpreting extensions of A within
A.
(V) Levy [ 10] showed that (the relativization to V of) the complete
reflection schema of ZF (which may be used instead of the replacement
schema in axiomatizing ZF) holds for A 1t formulas (in t.he hierarchy of
formulas developed in [ 101 .)
0.2. Results
The main result of this paper is t~le converse of (1):
(2) If o is a theorem of ZF, then the relativization to V of o is provable

in A*.
This, in conjunction with (1), shows that A* and ZF give the same in-
formation about sets, and answer; the question of Levy mentioned
above. It should perhaps be remarked that the proofs of both ( l ) and
(2) are finitary, Recall that Ackermann originally stated that the re-
placement schema of ZF (relativized to V) can be proved in A. Notice
that (2) establishes this with A replaced by A*, i.e. assuming the axiom
In trod~tc tion 193
of regularity for sets. Since A* is consistent relative to A in much the

same way that ZF is consistent relative to ZF without regularity, this
shows that Ackermann's statement is correct in the universe of "regular"
sets and classes. 1 do not know whether the assumption of regularity is
necessary (see question 4.24 and preceding discussion there).
The proof o~" (2) is contained in Section 3. Actually the following
stronger assertion is proved (Corollary 3.11 ).
If F is any finite set of formulas involving only ~ and having one flee
variable x, and q, is one such formula, then
(,2) A * I-- , I , ( V ) --, V ' v ' ( q , ( V " ~ A V ~ V' A
A /~ AXE V lO v ~ 0 v'] ).
~-~ E !:
•This is stronger than (1). Sections 1 and 2 provide necessary prelimi-

naries. In these sections, a n-amber of ki~own results which will be used
in Section 3 are collected. In Section 1 it is also shown that a certain
weak form of flae schema of replacement holds in A* (theorem 1.10).
This result does not seem to have been observed before, and plays an
important role ill Section 3. Section 2 contains some general observa-
tions concerning what can be proved in A* about the class of all ordi-
nals which are sets. These observations enable us to apply in A* certain
theorems which are known in Z. In particular, we apply Grewe's ~-esult
on natural models of A [6, 7] (see theorem 2.8 of this paper; for the
particular version we need, see theorem 2.12). it should be mentioned
that the author's original proof of (2) made use of Grewe's method of
interpreting theories in A, but that he subsequently found the simpler
proof presented here which avoids use of these interpretations.
Ic Section 4 we prove some additional results related to Ackermann's
theory, and mention some questions that remaia open. We prove (I)
A* is essentially not finitely axiomatizable over the theory whose
axioms are the instances of A2. (II) The theorems of A * (which is ob-
tained from A~o by adding for each V n the regularity axiom A5 with V
replaced by V n) involving only e and V 0 are exactly tl~e theorems of
A* (when V is identified with V0). This shows that A * is an extension
194 A ckermann Y set theory equals ZF
oF A* only in that it has some new individual constants. (III) If (R/3,~ e)

is a proper second order e x t e n s i o n o f ( Ra, e) (Re < 2 RI3 and c. E/3), ~"
then (R/3 + 1, e, Ra) is a m o d e l o f A*. We m e n t i o n a related question
o f Levy and Vaught (unpublished% Can there be ordinals a E ~ less than
the first inaccessible cardinal such that Ra -< 2 R~? A partial answer is
k n o w n (Tharp [ 24] ; see also Reinhardt [ 15a] ), namely, if the axiom o f
constructibility (V = L in the n o t a t i o n o f GOdel [ 5 ] ) holds, then ~ is
2nd order indescribable (in the sense o f Hanf and Scott [ 7 ] ) and hence
inaccessible. We m e n t i o n here only t w o o f the questions raised in Sec-
tion 4. (A) Suppose (Rt~+l , e, R~) is a m o d e l o f A. Is there an ordinal 3'
(a < 7 ~< t3) such that R e .<2 Rv ? (B) Is A + Z finitely axiomatizabIe
over A? It should be n o t e d that (assuming the consistency o f ZF) the
consistency o f A (when suitably formalized in A + Z) cannot be proved
in A + Z; thus k n o w n m e t h o d s o f obtaining a negative answer fail. A
positive answer, however, seems quite unlikely.
Section 5 has t w o parts. 1'he first part e x t e n d s the main result of
Section 3 to certain extensions o f Z and A. The second contains an
analysis o f the intuitive ideas behind A c k e r m a n n ' s theory. This analysis
suggests the i n t r o d u c t i o n o f an e x t e n s i o n A + o f A. Further extensions
o f A + and related axioms for set theory will be studied in a n o t h e r paper.
These axioms are c o m p l e t e reflection principles which seem natural and
plausible.
0.3. Set theoretic notation

"fie use the usual n o t a t i o n for m e m b e r s h i p (~), inclusion, ( c ) , un-
ordered pairs ({x, y } ), ordered pairs ((x,)'>)9 power set (Sx), union
over a set ( u x), cartesian p r o d u c t (x x y). difference (x -- y), and
union o f two sets (x u y). We write { t : P(t) } for the set of all t such
that P(t) (when such a set exists), and { (x, y ) : P(x, 3')} for { t : there
are x , y such that t = (x,v) and P ( x , y ) } . We put Ex = {(u, v) : u E v
and u, v 6 x } ; we will write (x, e) instead o f ( x , ex). Functions are
taken to be sets o f ordered pairs in the usual way; Xy is the set of all
functions with d o m a i n x and range included in y. We will occasionally
w r i t e r : x --, y f o r f ~ Xy. I f x is in the d o m a i n of)'; the value o f f a t x
is written f ( x ) (we will also write l)c or Ix occasionally); if x is not in
the d o m a i n o f f , we put f ( x ) = 0. l f x , y are sets, we say x is (set theo-
lntroductio~ 195
r e t i c a l l y ) e q u i v a l e n t to y in case t h e r e is a f l m c t i o n ) " : x ~ .v w h i c h is
1 -- 1 and o n t o y . We say x is transitive iff y ~ z ~ x implies y ~ x. We
d e f i n e o r d i n a l s in t h e usual w a y as transitive sets w h i c h are well o r d e r e d
u n d e r t h e e r e l a t i o n : the n a t u r a l n u m b e r s are the o r d i n a l s w h i c h in addi-
t i o n are well o r d e r e d by t h e c o n v e r s e r e l a t i o n . We let w be t h e set o f all
n a t u r a l n u m b e r s : t h e e m p t y set 0 is t h e first n a t u r a l n u m b e r . If u, 13 are
ordinals, c~ +/3 is t h e o r d i n a l s u m o f u and/3, a n d we w r i t e a < / 3 for
~ ~, o~ </3 f o r o~ </3 o r ~ =/3. An o r d i n a l is called a c a r d i n a l in case it
is n o t e q u i v a l e n t to a n y smaller ordinal. A f u n c t i o n w h o s e d o m a i n is a
n a t u r a l n u m b e r is called a finite s e q u e n c e . I f / ' i s a f u n c t i o n , t h e n
.f()i> = ( . f ~ { (.x,.fx> } ) u {O,', .v> }. If a is an o r d i n a l a n d t h e r e is a f'anc-
tion .f s u c h t h a t for every/3 < a a n d an3 t, t ~ ,f/3 iff for s o m e 3' ~ 13.
t c .t7, t h e n we w r i t e Ro~ =/ix (= sets o f :~'ank < c~). We w r i t e R a + n = R ( a + n ) .
0 . 4 . 5 ) ' n t u c t i c a l n o ta tioll
We will be c o n c e r n e d w i t h first o r d e r language w i t h variables
v o, v I . . . . . t h e usual !ogical s y m b o l s =, ~ , A, a n d A, p r e d i c a t e s y m b o l s ,
o p e r a t i o n s y m b o l s , a n d individual c o n s t a n t s , t o g e t h e r w i t h t h e usual
rules o f f o r m a t i o n . O u r p r i m a r y c o n c e r n will be w i t h a l a n g u a g e w i t h
o n e b i n a r y p r e d i c a t e s y m b o l e (we w r i t e e~--e o r e = ~ if it will c a u s e
n o c o n f u s i o n ) , a n d o n e individual c o n s t a n t V. In t h e a b s e n c e o f indica-
tion to t h e c o n t r a r y , (9, ,,, q, will be f o r m u l a s , a n d ( t o simpli;'y n o t a -
t i o n ) ,c, y , z,/3 . . . . will be r e s p e c t i v e l y t h e variables v 0, v 1 , .... In t h e
usual w a y , we will w r i t e -(9/~ ~ f o r t h e c o n j u n c t i o n o f (9 w i t h ~', ~ (3
for t h e n e g a t i o n o f ®, etc. Also we will w r i t e A u E v(9 f o r A u ( u ~ v -~
0 ) , u. v ~ w for u ~ w A v e w, etc. I f O a . . . . , 0 . _ 1 are f o r m u l a s , t h e n
/Ik ~bn, will be t h e cot~iunction 0 0 A...A ( 9 , ~ : similarly f o r t h e dis-
m E n
junction Y4 ,1, m . If u is a variable o r individual c o n s t a n t , a n d v is a

.tl ~ 11
t e r m , b,: O(,,
- " ) we u n d e r s t a n d t h e f o r m u l a o b t a i n e d f r o m O b y r e p l a c i n g
all free o c c u r r e n c e s o f it b y v, a n d c h a n g i n g t h e b o u n d variables o f ,I, t o
avoid collisions if n e c e s s a r y (all o c c u r r e n c e s o f an individual c o n s t a n t
are free o c c u r r e n c e s ) . We m a k e t h e c o n v e n t i o n t h a t if x 0, x l , -.., x n - l
are terms., O ( x 0, ..,, x , I ) will be t h e s i m u l t a n e o u s substittation o f
x 0 , •.... ~,z-l f o r v 0, ..., o , l r e s p e c t i v e l y (again c h a n g i n g b o u n d
variables if n e c e s s a r y ) . By t h e f o r m u l a O ~ we u n d e r s t a n d t h e f o r m u l a
196 Ackermann 's set theory equals ZF
o b t a i n e d f r o m 0 by relativizing all quantifiers to ~ (A uO goes to

Au(~0(u) ~ ~ ) ) ; i f u is a term, we will write 0 u for 0 tx ~ ' ~ . We will say
t h a t 4 is an e-formula if the o n l y non-logical c o n s t a n t o c c u r r i n g in 0 is
e; if we wish to e m p h a s i z e t h a t a f o r m u l a m a y involve b o t h e and V, we
will refer to it as an e, V-formula.
The following n o t a t i o n is n o t s t a n d a r d but will be very useful in
sections 1 - 3 . If 0 is a f o r m u l a and o a variable or an individual con-
stant, t h e n UoO is the f o r m u l a
O A Ao'(O(o°,) -* o = o')
w h e r e o' is the first variable distinct f r o m o not o c c u r r i n g in 69.

N o t e : v is n o t b o u n d hi UoO. UoO has all the free variables ~ has (as
well as v), and asserts that 'o is the u n i q u e v such that O'. W" will write
V!uO for V u U u • , and V!u ~ oO for V ! u ( u E o ^ O) (where u is any
variable and o any term).
If X; is a set o f sentences, and o is a s e n t e n c e , X; t- o will m e a n o is
derivable f r o m X; (in the usual first o r d e r logic with equality (=)). We
will write t- o ip case 0 i- o. X; is a t h e o r y if for every s e n t e n c e o ( o f
the language o f ~), if ~ I-- o t h e n o ~ 1~. If O is not a sentence, we will
u n d e r s t a n d Y. i- O to m e a n that if o is any universal closure o f ( ~ , t h e n
I-- o. If S and T are theories, S I-- T v m e a n s that for every o ~ T,
S I-- o v .
If T, S are sets o f sentences we will write T + S for the t h e o r y
{ o : T u S I- o } ; if o is a s e n t e n c e we will write T + o for T + { o }.
The set t h e o r y Z o f Z e r m e l o ) f o r m u l a t e d in a language with one
b:~aary predicate s y m b o l e ) has the axioms o f pairing, p o w e r set, sum-
se t, f o u n d a t i o n , and infinity (for f o r m u l a t i o n s o f these a x i o m s see for
example Fraenkel and Bar-Hillel [41 ) and the c o m p r e h e n s i o n schema
VzAt(t~z ~ ; t~xÔ)
where 0 is any f o r m u l a in w h i c h z does n o t occur. Zermelo-FraenLel set

t h e o r y ( Z F ) has in a d d i t i o n the r e p l a c e m e n t schema
A u V!t~o -* V z A t ( t E z ~ V u E x~o)
Introduction 19 7
where z does not o c c u r ;,n ~p. Q is the axiom
A x V y V r ( R n O ' , r) Ax ~ r) ;
Z + is the t h e o r y Z + Q ( R n ( y , r) is a f o r m a l i z a t i e n o f r = Ry; for an ex-

plicit definition see Chapter !, p. l ). We also will w'ant to c o n s i d e r a
w e a k e n e d version Z - o f Z; Z is the same as Z except t h a t the com-
prehension schema o f Z is replaced by the weak c o m p r e h e n s i o n schema
VzAt(t~z~-,t$,x AO y )
( w h e r e z ~ y and z does not o c c u r i~: O). Z 0 is Z w i t h o u t f o u n d a t i o n .
0.5. Semantical notation

We will write '.)l t-:- ¢ [ x ] in case ?( is a structure, ~0 is a f o r m u l a o f a
language appropriate to ?1, x is an assignment to the free variables o f ~0
taking values in the universe o f ?1, and ~0 is satisfied in '~1 by the assign-
m e n t x. If o is a sentence v~e write ?! ~ o if for every assignment x,
',~1 ~ o[x].
If ?l, ~ are two structures of the same type, ',~1-- ~ will m e a n t h a t for
all sentences o (appropriate to that t y p e ) ?11= o iff 2~ l= o, and Pl < '.~
('.)1 i~ an e l e m e n t a r y s u b s t ; u c t u r e o f ~ ) will m e a n t h a t for all f o r m u l a s
® (appropriate to that t y p e ) and all assignments x to the free variables o f
O taking values in the universe o f ')|,
,.~( ~ o [x t i f f ,~, ~ ~ [ x l .
(See Tarski-Vaught [ 19] for discussion o f these semantical notions;

there 'k~ is an arith:netical e x t e n s i o n o f ')l' is used for ".~l is an elemen-
tary substructure o£ ~ ' . )
198 Ackermann "s set theory equals ZF
,Section 1
In this Section we define a number of e-formulas we will have need

of i~l later chapters, discuss the absoluteness of these formulas, and col-
lect a number of basic results of the theory A.
1.1. S o m e e - j o r m u l a s . We will have need to refer to the following e-for-

mulas, These are taken directly from Levy-Vaught [ 111, p. 1047.
Definition 1.1. a. x g y is the formula A t ( t ~ x -* t ~ y ) and is read

"x is included in y'. We will frequently omit the dot.
b. Cp(x) is
(Ay, z)(z E y A.v E x ~ z E x)
'x is complete (or transitive)'.
c. So(x) is
Cp(x) A ( A y , z ) ( z ~ y A v ~ x ~ z E x )
'x is supercomplete (or strongly transitive)'.
d. Ord(x) is
Cp(x) A (Ay, z, w 6 x)iy ~ y A ( y E Z, V F = Z V Z E ) ' ) A
/~(w~y~z~w~Z)IA Au[u ~ x ^ Vt(t ~ u)-*
Vt(t E u A ~ V s ( s ~ t ,^ s ~ tt))] /',
^ (Au, v ~ x ) ( V w C_ x ) A t ( t ~ w ~ t ~ u A t ~ v)
'X is an ordinal'.
e. Db(x, y, z) is
At(tEz ~ ~ t=xvt=y)
'z is the doubleton { x, y } '
f. Oc(x, y, z) is
(Vv, w G z)(Db(x, x, o) A Db(x, y, w)A Db(v, w, z))
'z is the ordered couple (x, y ) ' .

Section 1 199
o Rl(x v , z ) is
V v(Oc(x, y, v,' ,, v E- z)
'x is in the relation z to y ' .
h. Rn(x, y )is
Ord(x) A Vz R n k ( x , y, z), where
R n k ( x , y, z ) is
A u ( u ~ x ~ V ~ Rl(u, v, z)) A
AAu, o{[uE.x'ARI(u,v,z)] vlu =xAv =y]
-~ A t [ t e v *--~ V u ' , v'(u' ~ u A RI(u', v', Z)A t C_ v')] }
Rn(x, y ) is read 'y is the sets o f rank less than x'.
i. lira(x) is
Ord(x) A A t ( t E x -~ V u ( t ~ u 6 x))
'.,," is a limit ordinal'.
1.2. Absoluteness. We will occa-sionally need to observe t h a t certain e-

formulas are absolute, so we begin w~th some l e m m a s on absoluteness.
Definition 1.2. Let ~f be a t h e o r y and let U ~-e a formula o f the language

o f T whose o n l y free ,;ariable is x. ~, f o r m u l a o is said to be absolute
with respect to U over T i f f
T I - U v 0A Uv I ^...A U ~ _ l - + l O + - - , O U] ,
w h e r e the free variables of 0 are e.mong v 0 . . . . . On-l"
If V is an individual constant, then instead o f saying absolute with re-

spect to the f o r m u l a x ~ V, we will say absolute with respect to V. We
witl say that 0 is absolute w h e n we m e a n that 0 is absolute w i t h respect
to Ux over T, where U is a u n a r y predicate n o t o c c u r r i n g in O, and T is
the t h e o r y (in the language a p p r o p r i a t e to 0 and U) whose o n l y axiom
is
200 Ackermann's set theory equals ZF
Ax A t ( U x ^ (t c_ x v t ~ x) -~ U t ) .
(Let us refer to this axiom as Sc U.)

O is absolute over A (resp. A*) will mean that O is absolute with respect
to V over A (resp. A*).
Lemma 1.3. L e t T be a theory, U a f o r m u l a o f the language o f T whose

only free variable is x.
a. I r a f o r m u l a ® is absolute with respect to U over T, and
T I- Uo 0 A..A Urn_ 1 A Vx O ~ Vx(Ux A O ) ,
where the free variables o f O are a m o n g v o .... , % - 1 , then V x 0 is abso-

lute with respect to U over T.
b. Suppose
T I-- A x A t ( U x A (t C_ X V t ~ X) -* U t ) .
Then the set o f f o r m u l a s I" absolute with respect to U over T contains

all atomic formulas, and is closed under negation, dis/unction, and
b o u n d e d quantification. That is, i f 0 and ~oare in I', then so are 6) A ~o,
"" O, VX(X ~ y ^ 0 ) , and V x ( x c_ y v 0 ) .
The proof of Lemma 1.3 is trivial; part b follows from part a.
Using 1.3 and the trivial facts
Sc U I- Db(x, y, z) A Uz -* Ux A Uy
Sc U I- Rl(x, y, z) A Uz ~ Ux z U y ,
one easily checks that
Corollary 1.4. The f o r m u l a s c_, Cp, Sc, Ord, Db, Oc, RI, Rnk, and lim
are absolute.
1.3. Theorems o f A and A*. Next we collect for reference a number o f

basic results of the theory A.
111eorem 1.5. (Ackermann) A t- Zo v, and hence A* I- Z v.

Section 1 201
Proof. See A c k e n n a n n [ 1 ]. ( E x a m p l e : p o w e r set a x i o m Take O = t c x;

A3 yields t c x e V ~ t e V, so A4 says the p o w e r set o f x is in V.)
R e m a r k . Since t- Oc(x, y , z) A Oc(x, y, z') ~ z = .-', Z I-- Vz Oc(x, y, z ),

and Oc is absolute over A, by 1.5 A I- x, y e V ~ Vz e V Uz Oc(x, y, z).
On the basis o f the class c o n s t r u c t i o n schema (A2) and the axiom o f

e x t e n s i o n a l i t y (A I) we see that
A b Vx Ux At(te x ~ t e VA O r d I t ) ) ;
this justifies i n t r o d u c i n g an individual c o n s t a n t c~, and assuming
A P Uc~ A t ( t ~ c~ ~ t e V ^ Ord(t)) .
(Recall that U x O ( x ) means 'x is the unique x such that O ( x ) ' . )
T h e o r e m 1.6.
a. H Rn(x y ) - ~ S o ( y ) ,
b. k- Rn(x. y ) ^ x ' ex ~ u ' ( R n ( x ' , y ' ) ^ y' e y ) ,
c. A 2 i - R n ( x , y ) A x C- V - ~ x C _ y ,
d. A t-- Ord (c~),
e. A t - A x ( O r d ( ~ ) - * x e a v x = ~ v a e x ) ,
f. A I - - A x e ~ v v~VRn(x,y),
g. A ~- A x < a ( O r d ( x ) ~ V y C V U y R n ( x , y ) ) ,
O
h. A* I-- Rn(o~, V).

All o f a - h are easily proved (Proofs o f some or all can be f o u n d in
Levy [ 8 ] , Levy-Vaught [ 1 1 ] , or G r e w e [6] .) We r e m a r k d t h a t part e
appears to be sensitive to the particular form~Jla c h o s e n for Ord, and
that this is the reascan for the clause
A u , v c_ x V w c x A t ( t e w ~ t e u A t ( ~ ~)
in the d e f i n i t i o n o f Ord.
202 Ackermmm "s set theory equals ZF
On the basis o f T h e o r e m 1.6 part g, we are justified in i n f r o d u c i n g a

u n a r y o p e r a t i o n s y m b o l R a n d assuming
A I-- x c_ a AOrd(x)-*- Rn(x, R x ) .
When it will cause n o c o n f u s i o n , we will o m i t the bar over the R.
T h e o r e m 1.7. (Levy).
a. (Non-definability o f ~)
AI--~[y~VAUs®(~,y)] ,
b. (Reflection principle for ~)

A I-y E V A (7,~)(ot, y ) ~ V x E a 0 ( x , y ) ,
c. (Reflection principle for V)

A* I - y E *V A l ~ ( V , y ) ~ Vz~Z V O(2,y),
d. (Upward reflection for ~)

A t-- y ~ VA O ( ~ , y ) ~ VX[~ ~ X O ( x , y ) ] .
Proof. We argue informally in A.

a. Assume y ~ V A Uo~ ®(a, y); apply'ing A4 to the formula
Vx(t ~ x A O(X, y ) ) we o b t a i n a ~ V. But t h e n a ~ ~,, which c o n t r a d i c t s
Ord(~).
b. Using the fact that A I--- A x ( O r d ( x ) -+ x ~ ~ v x = a v a ~ x), one
easily see:~ that (in A)
[®(o~,y)A "~ VX E 0~ O ( x , y ) ] ~ Ua[Ord(o0 A
A O(~,y)A "" V X E ~ O ( x , y ) ] .
F r o m this and a, b follows i m m e d i a t e l y .

c. follows by applying b to the f o r m u l a
Vz V r [ R n ( x , r ) Az ~ rA O ( z , y ) ] ,
and using the facts

Section 1 203
A* F- Rn(e~, V)
and
A t- A x ¢ c~ V y e V U y Rn(x, .v)).
d. The p r o o f is ;ike that for b.
Remark. One can easily see that replacing the schema A4 by the reflec-
tion principle for V yields an alternative axiomatization o f A* (see Levy-
Vaught [1 l l , p. 1057).
Corollary 1.8. Rn is absolute over A*.
Proof. It is enough to show that (in A)
1) x, y ~ V ^ Rnk(x, y, z) --, Vz' E V Rnk(x, y, 2')
as then we can apply L e m m a 1.3. It is i m m e d i a t e from the definition of

Rnk that
Rnk(x. y , z) ^ x' ~ x r, Rl(x', y', z) --, y' ~ y ;
using this it is easy to check that
Rnk(x,y,z)^x,yE V ~ V z ' C_ V R n k l x , y , z ' )
(just by taking z' = z n V. It is necessary to observe that Rl(x~ y, z) ^

^ x , y 6 V --, Rl(x, .v, z')). Now applying T h e o r e m 1.7c we get
Vz'C_ V R n k ( x , y , z ' ) - , V v ~ V Vz'C_ v R n k ( x , y , z ' )
so evidently 1 ) follows.
Corollary 1.9. A* F- (Z + )v.

Proof. Recall that Z + = Z + Q. We are to see t h e n that
A* I-- [ A x Vy V r ( R n ( y , r) ^ x ~ r)] v . Since A* F- x ~ V
Vr(Rn(c~, r) ^ x E r), by the reflection principle (! .7b) we obtain
204 Aekermann's set theory equals ZF
A* 1-- x E V ~ V y ~ V V r ( R n ( y , r) ^ x ~ y). Using 1.6 (parts f, g), and

Corollary 1.8, the t h e o r e m follows.
Theorem 1.10. (Weak r e p l a c e m e n t s c h e m a in A*).

Let ~o be an e-formula with free variables among x, y, u, t. Then
A* t - x , y ~ V^ Au~ VV! tE V¢~
Vz ~ V Ate V [ t ~ z ~-~ Vu ~ x ~ o ] .
Notice that this differs from the r e p l a c e m e n t s c h e m a retativized to V

o n l y in that ~o is n o t relativized to V.
Proof. We argue i n f o r m a l l y in A*. Assume x, y E V ^ Art E V V! t ~ V ¢:

by A3 we also have t h e n x c_ V. Using the reflection principle 1.7c we
c o n c l u d e that for some z' ~ V,
I) x, y E z ' / ~ X C _ Z ' / ~ A U E Z ' V ! t 6 z ' ~ ;
byA3wealsohavez'CV. N o w use A2 to form z = { t ~ V : V u 6 x ¢ } .

Evidently z c_ z' E V (if u c x ^ t E V ^ ~0(u, t), t h e n by 1 ) we may
choose t' ~ z' c_ V so that ~o(u, t'); o u r h y p o t h e s i s on ~ assures that
t = t', so t ~ z'), so by A3, z 6 V. The t h e o r e m follows i m m e d i a t e l y
f r o m the definition o f z.
Remark. This t h e o r e m seems not to have been observed before. It will

play an i m p o r t a n t role in Section 3.
Theorem 1.11. Let T be a theory (in the language with e) such that
A * I T v , (i.e., for every sentence o, T t- o implies A* I- o V ) a n d let 0 be
any e-formula who,~.e only free variable is x.
N o w if O isabsolute over A* and valid in T (i.e., A* I- A x
E v [ o V ( x ) ~ , O(x)] a n d T ~ A x ® ( x ) ) , d w n A* I- ®(cx).
Proof. I m m e d i a t e f r o m tile fact that
A I- A x e V[O v (x) ~ O(x)] ,~, A x ~ V 0 v (x) ~ 0 ( o 0 ,

Section 1 205
which follows from (the contrapositive of) the reflection principle for
(l.7b).
Remark. tn Section 2, we will apply 1.1 taking Z + for T.
Theorem 1.12, (L~vy, Vaught). A n #lterpretation o f ~,* #z A is deter-

m i n e d by: the interpretation o f x ~ y is x E y, :he interpretation o f V~c ..
is V x . . . . and the intetT)retation o f V is R~.
Proof. See L~vy-Vaught [ 11 ].
Before proceeding to Section 2, we define some formulas which will

be useful later on.
Definition 1.13.
a. Ct(x, y, z) is
A t [ t ~ z ~-~ V x ' ~ x V y ' ~ y Oc[x', y', t)]
Ct(x, y, z) expresses 'z = x × y:.
b. G(x, y, z, w) is
A t [ t E w ,--> ( O c ( y , z , t ) v
Vy',z'[y'= v A OcLv', z', t) ^ t ~ x ] ) ]
G(x, y z , w ) e.~:presses 'w = x Q')'.
c. F ' ( x , y , z ) is
Vu(Ct(y, z, u)^ x ~ u) ^
V y ' ( y ' ~ y ~ V z ' E z Uz' Rl(y', z', x ) ) ,
F'(x, y, z) expresses 'x ~ .Vz.
d. Nn(x) is
Ord(x) ^ Ay c_ x [ V y ' ( y ' ~ y ) -* V y ' ~ v At ~ y (y' q~ t)] ,
Nn(x) expresses 'x is a natural number'.

206 Ackermann's set theoo' ecuals ZF
e. Ffx, y ) is
Vz(Nn(z) ^ F'(x, z, y ) ) ,
F(x, y ) expresses 'x is a finite sequence with values in y'.
Theorem 1.14.
a. The f o l l o w i n g are p r o v a b l e in A*:
1. x , y E V - ~ Vz~VUzCt(x,y,z),
2. x , y , z E V -+ V w -~ V U w G ( x , y , z, w ) ,
3. N n ( x ) ~ x ~ V ,
4. V z ~ V Uz A t ( t ~ z ~. ; N n ( t ) ) ,
5. y , z E V A F ' ( x , y , Z ) - > x E V ,
6. y c _ V A F ( x , y ) - - > x a V .
1:. Moreover, Ct, G, F', Nn, a n d F atw all a b s o l u t e o v e r A *.
Proof. (al, a2). First observe that
1) Ct(x, y, z) ^ Ct(x, y, z') ~ z = z ' ,
and
2) G ( x , y , z, w ) A G ( x , y , z , w')--> w = w' .
Now we will apply Theorem 1.5 (A* I-- Z v ). Since A* ~ x, y ~ V

-~ V z Uz Oc(x, y, z) assures that
A* b x , y ~ VA Vx' ~ x Vy' ~ y Oc(x', y ' , t ) ~ t ~ V ,
we can apply the lernma on absoluteness (1.3) to see that C~:(x, y, z) is

absolute. Now
Z t-- Vz Ct(x, y, z~,
so, using, 1 ), part a l is clear.

Using Lemma 1.3 again, and the fact that
Sectkm 1 207
A*t-teVÔc(v,' ,.,,' t ) ~ . v ' , - ' ~ V ,
G is seen to be absolute. Arguing as before and using 2), a2 follows.

(a3, a4) Sihce A t-- Ay' ~ ~ Vt ~ a (y' ~ t), it is clear that
A t- a c_ ,- . ~ Nn(x). Thus by 1.6e (comparability of ordinals with a)
A t-- Nn(x) -- x ~ o~, and a3 holds. Since all quantifiers in Nn are
bounded, by 1.3, Nn is absolute. Now a4 follows from a3 (using 1.5
and the abso!uteness of Nn) and the fact that Z t- V z A t ( t ~ z ~ Nn(t)).
(b) We have proved the absoluteness of Ct, G and Nn by the way.
Since A t-- x ~ V .,x RI(y', z', x) ~ z' ~ V and RI is absolute,
Uz' RI(y', -', x) is absolute. Thus using al and the absoluteness of Ct, F'
is absolute. Since Nn is also absolute, a3 enables us to apply 1.3 to ob-
tain the absoluteness of F.
(a5) By the definition of F',
t-- F ' ( x , y , z ) ~ Vtt(Ctty, z,u)Ax C u),
so a5 follows by al and A3.

(a6) Observe that
l J Z i-- Ax At3{F(.,c, y) ^ Rn(/3, y)/x lira(/3) - x ~ y ) .
It is easy to check that the formula in 1) is absolute: F, Rn, and lira are
absolutc (by a3, 1.4, and 1.8), A* I--- Rn(/3, y) -~/3 ~ y, and (by a3, a5,
a~ad the definition of F) A* t-- F(x, y) Ay ~ V - 3- ~ V, so Lemma 1.3
applies. Thus Theorem 1.1 1 applies and we have
A* t-- F(x, V)A Rn(c~, V)A l i m ( a ) - x ~. V ,
from which a6 easily follows (using 1.611).

On the basis of Theorem 1.14, we may introduce individual con-
stants ,~ and (for each n c ~ } ix n , and a binary operation symbol Val
such that
A* I- A t ( t ~ Co, , N n ( t ) ) ,
A*I-- A t ( t ~ A o ~ t ¢ t) ,
A * 1-- At(t~ A n + 1 *-+ t ~ A n v t = A,,),
A* I- x , y ~ V ^ Uz RI(y, z, x)-* V a l ( x , y ) = z .
We will omit the bar from ~ if no confusion will result.

Section 2 209
Section 2
In this Section we will use T h e o r e m 1.11 o f Section 1 to show that

certain sentences o f the f o r m ,I, (c~)are provable in A*. The s e n t e n c e s in
questio~ express k n o w n results, so we will merely sketch the proofs in-
formally in Z+, and take care to notice that the notions involved in the
t h e o r e m s are absolute.
2, 1. Semantic and syntactic notions in Z. It is well k n o w n t h a t all syn-
tactic notiol~s are expressible in first o r d e r arithmetic as properties o f
(ol o p e r a t i o n s on) natural n u m b e r s , so there is a natural way o f for-
realizing all syntactic n o t i o n s in Z using formulas o f the for:-, 5) R~ ,
where O is an e-formula. Since A* F- Rw 6 V, evidently every sv ,~
formula is absolute over A*. in particular, we m a k e the followir defi-
nition"
Definition 2. I. We understand ZF to be a formula o b t a i n e d by imitating

the informal definition (given in Section 0) of 'x is an axiom o f ZF'.
R e m a r k . Of course as it stands Definition 2.1 is not c o m p l e t e l y precise'.

On the otiier h a n d there is no d o u b t that it could be m a d e quite precise
by explicitly exhibiting such a fo~'mula, and also no d o u b t that the
effort w o u l d not be very rewarding: Definition 2.1 will be sufficiently
precise for our purposes.
T h e o r e m 2.2. a. Z F is absoh~te over A*.

b. I f o is an a~:ioni, o f ZF, then Z I- ZF(Ao).
It is also well k n o w n that there is an e-formula Sat(x, y, z) (formaliz-

i~.g (z, e> ~ y [ x ] ) such that
1 ) If v~ is an e-formula, and the free variables o f ~0 are included
a m o n g o0, .... °,7 - l, t h e n
Zt- [ ~ Val(a,A m,o,,,)lAF(a,z)

m E !l
(Sat(a. A , z) ~ soz )
210 A ckermann 's set theory c ( r~clt ZF
w h e r e a, z are distinct variables n o t o c c u r r i n g i~a so or Sat.

We w o u l d like to have the additional i n f o r m a t i o n that Sat is absolute
over A*. ]'his is quite readily obtainable, since in any o f the usual for-
mulas we w o u l d take for Sat, all quantificatiol, is b o u n d e d (by simple
functions o f z, such as S z or S S S z ) . F o r example, one possible defini-
tion o f Sat would have it c o r r e s p o n d to the following informal defini-
tion o f sati,,~faction Jn e-structures:
Definition 2..3. We say that y is satis[ied in (z, e) u n d e r the a~signment x

iff
2) a) y is an e-formula,
b) there is n E ~o such t h a t x ~ " z and the free variables o f y are
anaongt, 0, ..., on_ 1 (that is, for every variable u o c c u r r i n g free
in y there is m ~ n such that u is the ruth variabl,- (u = v m )),
c) there is a relation s b e t w e e n subformulas o f y and m e m b e r s o f
n z (so s ~ ~ × n z ) such that ( x , y ) ~ s and
d) for every a ~ nz, and all s u b f o r m u l a s q s ¢, so', o f ) , , and all
rn, n E to such that v m , v** o c c u r in y, we have
1. if ,1, = so ^ so', the', (,I,, a) E s iff (so, a) ~ s and (so', a) ~ s;
2. if ,I~ = ~ so, then (q,, a) ~ s iff (~o, a) ~ s;
3. if ,I, = A v m so, t h e n (q,, a) E s iff for all t E z, (so, a(t2)) E s;
4. if q' = o m E u n , t h e n ~,I,, a) E s i f f a ( m ) E a(n).
N o w all the n o t i o n s occurring in this definition are either syntactic

and h e n c e absolute over A* w i t h respect to V (such as 'u = v m ',
'q' = so ^ so", 'y is an e-formula', etc.) or else have been formalized and
s h o w n to be absolute in section 1. F o r example, ' a ( m ) E a ( n ) ' is forma-
lized by
1) Vu Vv (Rl(m, u, a ) ^ Rl(n, v , a ) ^ u E v ) ,
'for all t E z, (SO,a(t~)) E s' by
2) A t ~ z A b ( G ( a , m , t, b ) -~ RI(so, b, s ) ) ,
'a E nz' by F'(a, n, z), and s c_ ~ × '*z by
3) A t E s V m V f ( N n ( m ) ^ F ' ( j , n z) ^ Oc(m, j~ t)).
The formulas displayed in ! ), 2), 3) are easily seen to be absolute by the

use o f the l e m m a o a absoluteness (1.3) and T h e o r e m 1.14. Of course all
quantifiers over ~o or formulas preserve absoluteness, so (writing
~I,(.~, n, z) for the formula displayed in 3)) Sat(x, y, z) has the form
Secnon 2 211
V , I F'(x, n,z ) A % (y, n) ,\ Vs(,~, (s, n, : ) A
A A a l F ' ( a , n, z) -, ~ l (a, n, : , y , s)l )l ,
where F', ¢0, ~p, and ~1 are absolute. Now Sat is easily seen to be ab-
solute using Lemma 1.3 and the following fact:
4) A* H n , z ~ V A ~I,(s, n, z)-* s e V .
The latter is easily checked; obviously
A * H n, z ~ V A q , ( s , n , z ) ~ s~ V,
so applying the reflection principle over V to
As(q,(s, n, z) -~ s ~ V)
we get 4). Evidently we also have in A*
S a t ( x , y , z)A z 6 V ~ x , y 6 V .
Thus we are led to make the following definition:
Definition 2.4. We understand Sat to be an e-formula obtained by for-

malizing clause 2) of Definition 2.3.
Remark. This is the last vague definition, All remaining definitions are
given precisely in terms of Sat and ZF. In particular all that is needed in
Section 3 is that there are e-formulas Sat and ZF such that Theorem
2.12 holds.
The next theorem summarizes the properties of Sat so far discussed.
Theorem 2.5. a. A* ~- Sat(x, 3,, z)/x z ~5 V ~ x, y ~ V.

b. Sat is" a b s o l u t e o v e r A*.
c. I f ¢ is an e-form,d,~, a n d t h e f r e e ~ariables o f ~ are i n c l u d e d a m o n g
v O, ..., vn_ 1, the~l
212 Ackermann "s set theoo' equals ZF
Zt-'[ I~ Val(a,/~,n,Vm)^F(a,z)~(Sat(a,/~ z), ,¢:)

m ~n 'P'
where a, z are distinct variables n o t occurring #~ ~ or Sat.
2.2. S o m e theorems o f Z +. N o w we p r o c e e d i n f o r m a l l y w i t h i n Z + using

the notions i n t r o d u c e d above. A f t e r stating and proving the t h e o r e m s
we w a n t informally, we will again r e t u r n to the q u e s t i o n o f fomaaliza-
tion and absoluteness over A*.
Definition 2.6. lf~0 is a f o r m u l a with free variables a m o n g v o ..... Vn_ l ,

and a ~ h R . t and a is the u n i q u e e l e m e n t in R v such that (R.~, e)
~0[a (°)], we will say t h a t 0~ is definable in (R v, e) by the f o r m u l a
u n d e r the assignment a; in case a takes values in a set A we will say that
a is definable in (R~t, e) using p a r a m e t e r s f r o m A.
T h e o r e m 2.7. (Montague, L6vy, Vaught). I t ' - / i s an ordinal, a ~ 7, and

<Ra, e> -< <R'r, e>, then <Ra, e> is a m o d e l o f ZF.
S k e t c h o f proof. (See M o n t a g u e - V a u g h t [ 13]. Because we need to

k n o w 2.7 can be proved in Z ÷ we include here a sketch o f the proof.)
If 7 = # + 1, t h e n # is definable in W,7, e~ as the last ordinal. Hence ~ is
also the last ordinal in (Ra, e), i.e./~ + 1 = c~. Thus 7 must be a limit or-
dinal; this assures t h a t (Ra, e~ is a m o d e l o f Z, so it suffices to c h e c k the
r e p l a c e m e n t schema.
Evidently for ar~y c-formula O whose free variables are a m o n g x, y
we have
1) (RT, e, Ra) I= A x , y ~ V(® v ~ O),
so wc argue i n f o r m a l l y f r o m this schema. Also,
2) (RT, e, Re) N A x , y ( x E y E V - - , x E V ) ,
so we assume this axiom. We will show that the r e p l a c e m e n t a x i o m

holds relativized to V. N o w using 2), the h y p o t h e s i s o f the r e p l a c e m e n t
a x i o m relativized to V is o f the f o r m
Section 2 213
3) x, y E V A Au ~ x V! t E V cv (u, t, y ) ,
where the free variables of ~p(tt, t, y) are anaong tl, t, y. Now we fix x, y
satisfyipg 3). Applying 1 ) to ¢, we see
x,yE V A AuEx V!tE V~o(u,t,y),
SO
4) Vz[x, yez ÂuEx V!tez~o(u,t,y)] .
Now applying 1) to 4) (and using 2) again) we see
Vz e V [ x , y e z A A u e x V! t e z ~0v (u, t , ; ' ) ] •
consequently
5) VuCxs0V(u,t,y)~ tE z E V .
The comprehension schema of Z holds in CRy, e), so 5) assures that
VzEV(t~z~ Vu~x~0 v(u,t,y))
which is the conclusion of the replacement axiom relativized to V. Thus

(RT, e~ is a model of ZF.
The next theorem is Grewe's important partial result in the direction

of our main result (Theorem 3.1). It will be used in our proof of 3.1.
In understanding Theorem 2.8, it may be helpful to notice that
<R~, e, Ra) is a model of A iff a ~ t3 and a is not definable in iRe, e)
using parameters from Ra.
Theorem 2.8. (Grewe). Suppose that (~ is an ordinal, ~ E ~, and t~ is n o t

definable in (R/3, e) using parameters f r o m Ra. Then there is a 7 such
that ~ E 7 <- ~ and
(R~, e) -< (R'r, e).

I n d e e d , i f there is an ordinal 8 s u c h t h a t ~ E 6 a n d ~ is definable: #2

(Rt~, e) using p a r a m e t e r s f r o m R a , we m a y c h o o s e 3" to be t h e l e a s t s u c h
o r d i n a k o t h e r w i s e we m a y t a k e 3" = (3.
Sketch of Proof. (See Grewe [61 for a detailed proof.) Assume tim
hypothesis and let 3' be chosen as in the theorem. Evidently the choice
of 3' assures that the schema
1) x 0 . . . . . Xn_ 1 E V A O r d ( f ) A U 6 0 ( 5 ) ~ 5 EV
(where 0 is an e-formula with free variables among x o, x I . . . . . x,z._ l )

holds in (R't, e, Ra) (in other words, every ordinal definable in (Ry, e)
using parameters from Ra is in Ro~) because
if is an ordinal definable in (RT, e) in terms of parameters

from Ra ,
and 7 is an ordinal definable in (R~, e) in terms o f parameters
from R a ,
then 6 is definable in (Rt3, e) in re'ms of parameters from R~.
Evidently 7 cannot be a successor ordinal, so (R3,, e) satisfies the theory

Z and the axiom
Ax Vy Vr Ix ~ r,~, Rn(y, r)l .
Now in (R3,, e) we certainly have (for any e-formula O)
V u 0 - ~ V6U~5[ Vu V r ( R n ( 6 , r ) A u ~ r A O ) A
A" Vo'~ Vu Vr(Rn(6',r)AuErAO)]
and so the schema 1) assures that
2) XO' ""' X n - 1 E V A VuO "* V u E VO
(where 0 is a formula w~th free variables among x 0 ..... x n - l ) holds in

(R3', e. RcO.
Now from 2) it is easily seen and well-known that (Ra, e) -< (Ry, e),
as desired.
Section 2 215
2.3. Formalization m A*. Now we wish to state formal versions of

Theorems 2.7 and 2.8 and make some remarks on the absoluleness of
the formulas involved. For this pthpose we introduce some notation.
Sat is the formula of Definition 2.4.
Definition 2.9. (a) E ( x , y ) is the e-formula
Vr Vs[ Rn(x, r)/x Rn(y, s) ,',
/x A~. a(Sat(a, ~o,x) + Sat(a. ¢, v))].
E(x, y ) formalizes '(Rx, e) < (Ry, e)',

(hi ModzF(x)is the e-formula
As0{ZF(¢) -+ Va Sat(a, v~, x ) ) .
Modzv (x) formalizes '(v e} is a model of ZF'.

{c) S(x, y, z, u) is the e-formula
Vr Vn Vb[Rn(u, r)A AtIt ~ n)AG(b, v, n, z ) / x S a t ( b , y , t ) ] ;
here G(b, x, J~, z) is the e-formula introduced in section 1 which for-

realizes "b = x(',' }".
S(x,,7, z, u) formalizes'(Ru, e) ~ y[X(z0)] ,. Thus UzS(x, y z, u) for-
realizes 'z is definable in (Ru, e) by the formula y under the assignment
X ~"
(d) D ( x , y ) is the e-formula
V~, a, r[ Rn(x, r).,~ F(a, r) ^ UxS(a, ~, x, y ) l •
D(x, y) formalizes "x is definable in (Ry, e> using paran ;ers from
Rx'.
Theorem 2.10. E, Modzv, S, UzS and D are absolute over A*.
Proof, To see that E, Modzv, S, UzS, and D are absolute, apply the
lemma on absoluteness of section I, making use of the tbllowing facts:
1. Sat, Rn, ZF, and G are absolute.
2. A* I-- Z F ( x ) ~ x ~ V .
3. A* i - x ~ V v Sat(a, ~0, x ) ~ a , ¢ ~ V.
4. A* t-- Rn(x, r) ^ x E V --> r ~ V .
5. A* I-- S ( x , y , z , u ) ~ z ~ u .
N o w we state the formal versions o f T h e o r e m s 2.7 and 2.8.
Theorem 2.11.
a. Z + I- a ~ ~ ^ E(a, t3) --> Vr[ R n ( e , r) A MOdzF (r)l ;
b. Z + I- a E / 3 ^ " - D(a,/3)Ôrd(#)~ V¢,aE Ra V3'(~ ~ 3' A
^ E(~, 3') ^ [3' =/~ v U3'S(a, ~0, 3', 8)] ) ;
,c. Z + I- Rn(/3, x ) AMOdzF(x)-> o x f o r a n y a x i o m o o f ZF:
d. Z + t-- a, 3' c_ ~ A~ ~ 7 A E(~, 7) ~ A x E R e [ o R S ( x ) ~ ORV(X)]

f o r any e-jbrmula ® whose only free variable is x.
Here ~, [3, r, ~o.... are o f ~ourse suitably chosen distinct variables.
Proof. a. is just T h e o r e m 2.7, b. is just T h e o r e m 2.8, c. is an i m m e d i a t e

c o n s e q u e n c e o f T h e o r e m s 2.2 and 2.5 (c). and d. is also a c o n s e q u e n c e
o f 2.5(c).
Theorem 2.12.
a. A* t-- A x ( x ~ a A ~ D(x, a ) - V r I R n ( x , r ) A MOdzF(r)]) •
b. A*I-- x(xEa^~ D(x,a)~V¢,aERx V3'(X~3, A
A E(X, 3') ^ [3' = ~ v U3'S(a, ¢, 3', c~)l )) ;
c. A* I- M o d z F ( R a ) ~ o r s for each axiom • o f Z F ;
d. A* I- Ax, 3' c_ 0~(x E 3' A E(x, 3') ~ A t e RxlORX(t)~-=~ @R~(t)] ),

where O ( t ) is any e-)brmula whose only free variable is t.
Proof. Using T h e o r e m 2.10 (and fact 4 o f the p r o o f o f 2.10) o n e easily

Section 2 217
checks (using the absoluteness lemma of Section 1) that the formulas in

Theorem 2.11 (considered as properties of ~, that is, leaving/3 free but
universally quantifying all other free variables) are absolute over A*.
Thus Theorem 1.11 applies and we conclude that in A* o~has ti~e prop-
erties considered. This proves 2.1 2 c and 2.1 2 d. 2.12 b follows because
A* I- Ord(~), and 2.12a then follows from 2.12b.
Section 3
In this Section we prove our main result,
Theorem 3.1. For every sentence o (involv;ng only e), o is a theorem o f

ZF i f f ov is a theorem o f A*.
The implication from right to left was proved by Levy [8] (for a
sketch of the proof see Theorem 4.18); we will prove the reverse impli-
cation. We begin with an important lemma. Recall that U/3O says tha t/3
is the unique/3 such that O. In other words, that/3 is defined by O (in
terms of the parameters of 0).
l.emma 3.2. Let O be any e-formula. Then
A* t-- x ~ V A U/~@(/3, x) ~ "-" D(ot,/3),
provMed the free variables o f O(/3, x) are among/3, x. (Here D(a, #) is

the formalized version o f '~ is definable in (R/3, e) in terms o f param-
eters f r o m Rot' which was indicated in Section 2. )
Proof. Consider the formula D(ot,/3); it is of the form
I) V¢ ~ ¢o Va[ Vr(Rn(ot, r) ^ F(a, r)) ^ UotS(a, ¢, a,/3)1 ,
where F and S are e-formulas. F(a, r) is a formalized version of 'a is a

finite sequence taking values in r', and S(a, ~, a,/3) is a formalized ver-
sion of 'if the first variable of ~ takes the value a, and the remaining
variables are assigned according to the sequence a, then ~ holds in
(R/3, e)'; thus UrnS(a, ~o, ~,/3) formalizes 'a is definable by ~ in (R/3, e)
using parameters ai'. The formalizations we have in mind were indi-
cated in Sections 1 and 2 (1.13, 2.9(; for the remainder of the proof,
however, S can be any quaternary e-formula, and all we need to know
about F is that A* I-- F(a, V) -~ a ~ V. Thus if we replace F(a, r) by
F'(a, t) = F(a, r) ^ a ~ r we obtain a result independent of the formali-
zations of Section 2.
We now argue informally in A*. Suppose that 1) holds. Now (in A)
Section 3 219
F(a, V) ~ a e V, and ~ _c V, so choosing a, ~ as in 1 ) we see
2) a, ~ e V ^ UaS(a, ¢, a, t3).
N o w the h y p o t h e s i s in L e m m a 3.2 is
3) x e V^ U~O(3, x).
Evidently 2) and 3) imply
4) x, a, ~0 e X, ^ Us Vt~(O(¢~, x) ^ S(a, ~, a, ~ ) ) ,
w h i c h c o n t r a d i c t s 1.7 a. H e n c e (assuming 3)) 2) must fail, and conse-

q u e n t l y 1) must fail, thus proving the lemma.
R e m a r k . We might say that the point o f this p r o o f is that " d e f i n a b l e

inside implies definable o u t s i d e " . Notice t h a t the hypothes~s o f the
formal assertion says that O provides an e-definition of/3 in terms o f a
p a r a m e t e r f r o m V. We k n o w that there is no e-definition o f ~ in terms
of p a r a m e ' e r s f l o m V = Ra (1.7a), and it is a techpical point ( w h i c h
h o w e v e r requires some care) to see that ( u n d e r the h y p o t h e s i s ) if
D(a, ~) i.e. 'a is definable in <R~, e> in terms of p a r a m e t e r s f r o m
Ra = V', t h e n we can find an e-formula defining a in terms of param-
eters from V. Of course, the passage to the outside adds new param-
eters.
Definition 3.3. Let q, be any e-formula whose only free variable is v 0.

T h e n D F * is the set o f all universal closures of formulas o f the f o r m
X O, . . . , Xn__ 1
e V ^ ,I, (t3) ^ Ord(13) A U~O(/3) ~ t3 e V ,
where x .., X n _ ~ , v 0, ~3 are distinct variables, and O is an e-formula

whose free variables are i n c l u d e d a m o n g x o , ..., X n _ l , Vo.
We will refer to D F * as the s c h e m a on definable ordinals ( o b t a i n e d
f r o m ,I, ). If q, is A x ( x = x ) , we will write DF for D F * . We have allowed
finitely m a n y variab!es x m in O; using the pairing axiom for sets (which
220 Ackermann's set theory ~quals ZF
holds io A*), finite s e q u e n c e s o f e l e m e n t s o f V are r e p r e s e n t a b l e in V

(that is,
A[-AXo,...,xn_ 1 e V V y e V /'A [Vai(Y, Am)=Xm]),

men
so it is easily seen t h a t (over A) one variable x 0 (in a d d i t i o n to ~) w o u l d

suffice.
Definition 3.4. Let q, be a n y e - f o r m u l a whose o n l y free variable is v 0 .

For each e, V-formula O and each variable u, Q,I, u® is the f o r m u l a
V~U#( [q,(~)A Vu V r ( R n ( # , r) AU ~ r , \ O)] A

A .-- V/3' E /ji[q'(~') ^ VU V r ( R n ( / f , r) A U E r ,~ O)1 } ,
!
A
where ~, r, ~' are the fir~t three variables w h i c h do n o t o c c u r in O.

Q * u o is to be t h o u g h t o f as a quantifier; Q * u® expresses 'There is
a u such that O(u), i n d e e d there is a u n i q u e minimal ordinal/~ such t h a t
q' (t3) and some u o f rank less t h a n # satisfies O(u)'.
Notice that the free variables o f Q * u o e x c l u d e u but o t h e r w i s e are
the same as those of O.
Definition 3.5. Let q, be any e-formula whose o n l y free variable is v 0.

l¢or
l
each e-formula O, 0 9 is the eormula d e t e r m i n e d by the recursive
conditions
(U E o)Q = U ~ o , (~ o ) Q = ' ~ (O Q ) ,
(O V ~p)Q = ( o Q ) v (~0Q), ( V / 4 0 ) o = Qq'/t(oQ ) ,
where u and v are variables and O, ~0 are e-formulas. Of course O Q de-

pends on ,I, ; if necessary we will write OQ q' .
Notice that the free variables o f ®Q are the same as those o f O, and
t h a t o Q is an e-formula.
N o w we state the main l e m m a s that will be used in the p r o o f o f

T h e o r e m 3.1.
Section 3 22 I
L e m m a 3.6. L e t ee be an e - J o r m u l a ( w h o s e oi~ly f r e e variable is v 0 ) smh

thot
A* 1-- q , ( ~ ) .
T h e n j o r e v e r y e-jOrmtda (9 w i t h .free variables x o ..... x,7__] w e have
A * + D F 'v I- x 0 .... , x l ~ V - - , [O v ÷ ~ -6)O'I'] .
( R e c a l l t h a t ~ is an indn, i d u a l c o n s t a n t i n t r o d t t c e d so t,,~at
A* I- A x ( x ~ ~ ~ ~ O r d ( x ) A x e V).)
I x m m a 3.7. L e t q, be an e-fpr/,zttla w h o s e o n l y f r e e ~.ariable is v o . T h e n

f o r each i n s t a n c e o o f D F 'v w e have
A* + ~ o h V/3[a ~/3 ;', q'(~) A ~ D(~, ~-~)l •
F o r a g l i m p s e o f t h e role L e m m a 3.7 will play, observe t h a t a ~/3

A "" D(a,/3) o c c u r s in t h e h y p o t h e s i s o f G r e w e ' s T h e o r e m 2.8.
P r o o f o f 3.7. Le~nma 3.7 is really a c o r o l l a r y o f L e m m a 3.2. T o see t h a t
3.7 follows f r o m 3.2. observe t h a t if o is an instance o f D F * , t h e n (with-
o u t loss o f g e n e r a l i t y we m a y a s s u m e ) --- o is t h e existential closure o f a
formula
1) x ~ V A q; ((3) A Ord(tS) A U~O(~, x ) A t5 ¢ V ,
w h e r e tp is an e - f o r m u l a a n d eyeD; free variable o f ~p(t3, x) is {3 or x. N o w

we a s s u m e 1) and argue i n f o r m a l l y in A*. A t h e o r e m o f S e c t i o n 1
(1.6e) assures that
O r d ( ~ ) n/3 q~ V --* ol E 13A~ = ~ ,
and o f c o u r s e by the r e f l e c t i o n principle 1.7a,
x~ V ^ Ut~O(t3, x ) - * a :~ ~ ,
so t h e c o n d i t i o n a 6/3 is satisfied. (Incident~tlly, this is t h e o;'ay p o i n t

where we make use o f the clause 'Ord(/3)' which appears in the defini-
tion o f D F * . ) E v i d e n t l y the Lemma 3.2 applies, so ~ D(0~,/3) and we
obtain the Corollary 3.7.
Proof of Theorem 3.1 from 3.6 and 3.7. Choose xp to be
w I ( t
A x E x[ ~ D(x , x ) -~ V r ( R m x , r) A MOdzF(r))] . ,;
By Theoiem 2.12a, we see that A* i-- q'(a), so both lemmas (3.6, 3.7)
apply. The essential point here is that a is quite well behaved in the
e-part o f :~he theory A*, even though (fc, r example) Zennelo's theory Z
does not hold for the classes of A. Evidently by 3.7 and our choice of
q', for each instance o of DF* we have
A* + ~ o k- MOdzF (Ra)
and consequently by theorem schema 2.12c for each axiom o 0 o f ZF,
1) A*+~o I-o~ .
On the other hand, using 3.6 and the weak replacement schema 1.1 O,
we see that for each axiom o 0 of ZF
2) A*+DF* I-o v "
indeed (we now argue informally), by 1.5 it is sufficient to show that

the replacement schema holds relativized to V. Evidently this amounts
to
3) x , y ~ VA A s ~ V V t ~ V[~oV (x, y , s, t) ^
A A t ' ~ V(,~v (x, y, s, t')-+ t = t')] -~
-~ V z ~ V A ~ V ( t ~ z < . ~ Vs~x~oV(x,y,s,t)).
Since all quantification in 3) is relativized to V, we see using 3.6 that

Section 3 223
replacing ~0v by ~0Q at each o f its occurrence~ yields a f o r m u l a equivalent

to 3). Since ¢0 is an ~-formula, 3) b e c o m e s an instance o f the weak re-
p l a c e m e n t schema. This c o m p l e t e s the p r o o f o f 2). F r o m 1) and 2) it
follows that for each a x i o m o o f ZF,
A* t- o v ,
w h i c h is the assertion o f T h e o r e m 3. !.
We now proceed to the p r o o f o f Lemma 3.6. The basic idea o f the

p r o o f is co,~tained in a s u b l e m m a :
L e m m a 3.b. Let q, be an e-formula (with one flee variable v o ) such that

A* t- q~(~). Then
A* + D F * I--x 0 .... , x n _ 1 E V - ~ [ V u E V O +-~ Q ~ ' u O ] ,
where @ is any e-formula and the free variables o f O are among

X O , """ X n - I , and u.
Proof. Since q, will be fixed t h r o u g h o u t the proof" we will write D F for

D F * , Qu "or Qu 'v , etc.
Without loss of generality, we may suppose that v o does n o t o c c u r in
O. Then Qu@ is
w h e r e x is the e-formula
'Is (t3) ^ Ord(~3) ^ Vr Vu[ Rn(/3, r) ^ u ~ r A O | .
Now we argue i n t b r m a l l y within A*. We are to assume x 0, ....

Xn_ 1 E V aI~d show Vu ~ V O ~ --~ QuO.
First suppese Vu ~ V O. In A*, Ra = V ^ Ord(~), and by h y p o t h e s i s
we also have , ' (a), so Vu ~ VO implies X(a). By the reflection prin-
ciple, VI3 E a X(~'). Using the class c o n s t r u c t i o n principle A2, we m a y
consider
A = {3 ~ o~ " ×(3)} .
N o w A ~ 0, so since O r d ( a ) t h e r e is a m i n i m a l m e m b e r ~ o f A . Using
basic p r o p e r t i e s o f ordinals, o n e easily sees t h a t this # satisfies
U/3[X(~) ^ A~' E ~ --, X(~')]
as desired. (The p r o p e r t i e s involved are: #' ~/3 ~ a -~/3' ~ a, and/3 ~ a ^

^ O r d ( y ) -~ # ~ y v/3 = y v y ~/3.) This d i r e c t i o n o f t h e e q u i v a l e n c e
does n o t t se t h e s c h e m a DF.
N o w s u p p o s e Q u O , i.e. V#U~3[×(lJ) n ~ V#' ~- ~ × ( ~ ' ) ] . We are to s h o w
Vu ~ V O . Fix j3 such t h a t
U/~Ix(~) A A~'~ ~ "" X ( ~ ' ) ] •
Evidently
x o, ..., Xn_ l ~ V A q'(~) A Ord(/3) ^
A U ~ I x ( ~ ) A A~' e fl "" ×(~')1 -~/~ e V
is an instance o f DF, a n d since x(~) ~ q" (~) ^ Ord(t3), t h e h y p o t h e s i s is

satisfied. T h u s # ~ V ^ X(~), so
~ V ^ Vr Vu[u ~ r ^ R n ( # , r ) ^ O l .
Since ~ ~ V, by T h e o r e m 1.6, R~ c_ Re = V, a n d we have Vu ~ V O as

desi~'ed.
This c o m p l e t e s t h e p r o o f o f L e m m a 3.8.
P r o o f o f L e m m a 3.6. L e m m a 3.6 is n o w i m m e d i a t e f r o m t h e s u b l e m m a
(3.8) by i n d u c t i o n o n f o r m u l a s . We discuss o n l y t h e case w h e r e
(9 = Vu,p. By i n d u c t i o n we m a y a s s u m e
1) Xo, .... Xn_ l , u ~ , V È !~v ~ , p Q ] ,
w h e r e x 0, ..., x n _ l , u are t h e free variables o f ¢ , N o w

Section 3 225
XO . . . . . . ~:n- ~" 11.~ V-'* [ V u E V(tpV ] ~ VII E V(tpQ)]
(by the induction hypothesis), and
-~ [ Vu ~ v ( ¢ Q ) , ,, Qu(¢Q)I
(by 3.8!, alld hence (by the definitions of 19v and O Q) we have
A'0, ...,Xn,_l E V ~ [( Vlttp~ ] ~ (V/t~p) Q] ,
which completes the proof for the case ® = Vu~0.

This completes the proof of 3.6, and he!ace Theorem 3. ! is proved.
Remark. It is perhaps worth remarking that on the basis of the theory T

obtained from A* by adding the axioms of Z and the axiom
Ax Vy Vr(Rn(y, r) A ,," ~ r) we can obtain a sharper result than 3.6,
namely
T + DF I- A x e V[6) v +--~-~9] .
This is because in T
v ~ V ,\ VxO(x, y) -~ Qx-O(x, y) ;
since o contains -mlimited quantifiers, the proof appears to make quite

essential use of the comprehension schema of Z (to form
{/3' e ~3 : V r V x ( R n ( f J ' , r ) ^ x e r ^ ® ) } ) .
Analysis of the proof of Theorem 3.1 shows that we can obtain a

slightly stronger version.
Theorem 3.9. For every e-formula O (whose only free varhlble is x)

there is an e-formula ~3 (whose only free variables are x and y) such that
A t-- V y E R~ /kx E Ra[®RC~(x)~--* ~ ( x , y ) ] .

226 Ackermann 's set the,~o' equals ZF
In view of the weak replacement schema 1. I0, Theorem 3.8 imme-

diately yields Theorem 3.1.
Remark. I do not know whether it is possible to eliminate the param-

eter y from O. In the abstract [ 15 ], it is mistakenly claimed (Theorem
3. o~ [ 15] ) that the proof of Theorem 3.1 yields this stronger result.
Theorem 3. of [ 15] should be replaced by the above Theorem 3.9.
Proof of Theorem 3.9. In view of the fact that there is an hlterpretation

of A* in A in which Rt~ plays the role of V (Theorem 1.1 2), it will suf-
fice to, prove the theorem in A*. We will need the following:
I.emma 3.10. L e t 0 be an e-formula with one free variable x, and sup-

pose there is a sequence @m " rn ~ n) o f e-formulas (each with free
variables x a n d y ) such that
A'l-- fit' VyEVAxEV[O v~.:] .
tn ~ ti
Then there is an e-form,da gO (with free variables x and z) such that
A*k VzeVAxeV[O v ~)1 •
Proof. Since
A*P-VyEV W A x E V [ O v ~ >%nl ,
m E n
evidently
A'P- Vy~VVuEVAxEV(O v, ;
~-~ W [ u = a,,, ^ ~., (x, y ) ] ) ,

m E n
so if we take (9(x, z) to be
Vu, y ( O c ( u , Y , Z ) ^ W [u =A m ^ % n ( x , y ) ] )
m En
Section 3 227
we easily obtain the c o n c l u s i o n o f the lem h a .

We n o w seek to satisfy the h y p o t h e s i s o f L e m m a 3.10. The idea is to
find (for each ® m e n t i o n e d in T h e o r e m 3.9) a f o r m u l a qs w h i c h we m a y
use in L e m m a 3.6 to obtain an e-formula @e such that
1) A * + D F ve D A x ~ - V [ O v ~ - ~ ® e ] .
Thus there will be a finite n u m b e r n o f instances crm o f D F * f r o m

w h i c h the c o n c h f 1 ) fellows. Wc wil~ c h o o s e ,Is so that we also
have, for each m . i ~ 2, formulas %,z ~ (with free variables x, y ) such
that
2) A* + "" o n I-- W Vy E V A x ~ V[O v s0;m1 .

i~2
The h y p o t h e s e s o f L e m m a 3.10 t h e n follow (taking the n of 3.10 to be

2n + 1, and setting ¢2n = ®Q), since
A*I--( W ~Om)V /~ om
m ~ n m ~ ~t
Thus to c o m p l e t e the p r o o f o f T h e o r e m 3.9, we n e e d o n l y find such a

q,, and prove 2) for it.
Choose ,P to be the c o n j u n c t i o n o f the formulas
A x ' i x ' ~ x/x ~ D(x', x) -+ V~0, a e Rx' V3,(x' ~ 3' A
A E(x', 3') A [ 3, = x v U3,S(a, ~, y , x ) l )1 ,
A x', 3,(x', 3, c x ^ E(x', 3,) -+

At E Rx'[ gRx' (t) ~+ ®RT(t)]).
Here D(x', x) formalizes 'x' is definable in (Rx. e} using p a r a m e t e r s

f r o m R x " , E(x', x ) formalizes R R x ' , e~ is an e l e m e n t a r y s u b s t r u c t u r e of
(Rx, e)', and UyS(a, ~0, 3,, x ) formalizes '3' is definable in (Rx, e) by the
f o r m u l a ~, using p a r a m e t e r s ai'; 0 is the f o r m u l a o f T h e o r e m 3.9.
228 Ackerman,Fs set t h ~ o , equals ZF
By T h e o r e m 2.12 o f Sectiun 2 it is clear that A* I- qt(a), so we m a y

use q, in L e m m a 3.6.
To show that 2) holds, we m a y suppose t h a t "-~ %: is ( t h e existential
closure of)
l) y E V ^ ~I,(~) ^ Ord(/3) ^ U/30 m (B, y ) ^/3 ¢ V,
where the free variables o f O,n ([3, y ) are/3 and y. We will argue infor-
mally within A*. Assuming 1 ), e v i d e n t l y L e m m a 3.2 applies, se we have
o~ ~ ~ A ~ D(e~,/3). We c o n c l u d e using q~(~) that there are ¢, a, and "~
such that ~, a ,.=-V and either
U T S ( a , ~p, 7 , / 3 ) A AX E Rc~[® R°~ ~ O R~' ]
or else
) = / 3 A A X E Ro~[® Ra ~ O R~t ] ,
so that eithe~-
2) t \ x e: R a [ O Ra ~ V/3 V T ( 0 m (/3, y ) A S(a, ~o, q¢, 13) ^ O R7 )]
or else
2') A x ~ R a [ O Ra , , V/3(0,n (t~,y) ^ o R a ) ] .
Evidently a does not o c c u r on the right h a n d side o f either 2) or 2').

N o w we can collapse the three p a r a m e t e r s ~0, a, y into one p a r a m e t e r 3'
(for example using Val). Thus we obtain formulas ¢,,~ i (i = 0,1 ~ such ,tllat
A* -t-~ o
m
t-- W V y ~ V A x E V [ O v , , e r ai 1
i
as desired. This c o m p l e t e s the p r o o f o f T h e o r e m 3.9.
Corollary 3.11. Let F be a finite set o f e-Jbrmulas with one free variable
x, and q, one such formula. Then
Section 3 229
A* l-- ,t,(V) ~ VV'(q,(V') ^ V ~ V' A

^ ~ A x ~ V [ O v ~ ,~0 v ' ] ) .
® el:
Proof. By Theorem 3.9. the existential closure of
1) ( /}{k . I ' o e V ) A /}~ A x E V I ® V ( x ) ~ " + ~ 3 ( x , y ~ ) ]

~@F OEF
is provable in A* (where e is a certain e-formula with free variables

%, o 1 ). Now using the upward reflection principle (1.7d), we see that
2) VV'01,(V')AVEV'^ /~ Ax~V'tOV'(x)~-~O(x,'; )])

OEF
follows in A* from 1 ) and the hypothesis q, (V). Thus the existential

closure of the conjunction of 1) and 2) is provable in A*; from this the
corollary follows at once.
Examination cf our proof of Theorem 3.9 shows that it gives an

effective way of proving in A the relativization to Ra of every axiom of
ZF. Tlms combining our results with Levy's results, we have a finitary
proof of
Corollary 3.1 2. ZF is c o n s i s t e n t if]" A is consistent.
Remark. Consider tlae theory A' obtained from A by replacing Acker-

mann's schema A J, by
1) A t I o ( t , x, 3') ~-" o v (t, x , y ) ^ t ~ V)I --,
-~ V z A t ( t e z ~ O(t, x, y ) ) ,
(.O|!¢ easily sees that 1) is equivalent to the schema obtained by adding

the ¢!ause
2) At ~ V[®(t, x, y)~-+ Ov (t, x, y)]

230 Ackermann •s set tlseor), equals ZF
to the antecedent of A4). Theorem 3.1 holds with A* replaced by A';

the proof is essentially the same. The main difficulty is in proving the
elementary results (e.g. those collected in Theorem 1.6, especially
f.g.h.); one must be careful to check absoluteness in order to apply 1).
Thereafter most of the results we prove for A* hold for A', with the
reservation that in cerlain schemata involving ~ one must add 2) to the
antecedent.
Section 4 231
Section 4
In 4.1 we give applications o f Levy's reflection principles to o b t a i n

i n f o r m a t i o n a b o u t the strength o f A*. In 4.2, we discuss natural m o d e l s
o f A*. S o m e questions are posed in 4 . 3 - 4 . 5 .
4.1. The strength o f A*

Definition 4.1. Let T1, T 2 be arbitrary theories.
a. T l is said to be finitely axiomatizable over T 2 iff there is a sen-
tence o o f the language o f T l , such that "1"2 + T 1 = T 2 + o.
b. T I is said :0 be essentially n o t finit~iy axiomatizable over T 2 iff
there is no consistent t h e o r y T' including ]'1 w h i c h is finitely axioma-
tizable over T 2 .
T h e o r e m 4.2. A* is essentially not Jbliteiv axiomatizab]e over the class

construction schema A2.
Proof. Suppose that T = A2 + O(V) includes A* ( w h e r e O is an e-for-

mula with one free variable x). We will show that T is inconsistent. One
easily shows that
A * [-- O ( V ) ~ VX(X E V ^ UX[Ord(X) ,,,
^ OtRX) a AX' e X "-- O(RX')I )
h e n c e as T t--- tg(V),
(1) T I- VX(X e V ^ U X [ O r d 0 , ) ^ O ( R X ) ^ A X ' ~ X ~ O ( R X ' ) I ) .
We m a y i n t r o d u c e an individual c o n s t a n t ~o for the X defined in 1): F o r

each instance ~(V) o f A2, we e v i d e n t l y have
T t- AX ~ o~ ¢(RX) ;
in particular, we have T 1-- ¢ ( R ~ o ) ^ O(RX0) , w h i c h we write symboli-

cally as
T(R~) ~ T(R~o).
Now by i),
T(V) I- V;~ ~ V[Ord(h) ^ O(RX)]
so evidently (replacing V by R?~0)
T(RX 0) I-- V~ ~ R;~0[Ord(~ ) ^ O ( R h ) ] .
On the other hand, the definition of ;~0 assures
T I-- A;~ ~ R;~0 [Ord(~) ~ ~ O(R;~)] .
Thus T is inconsistent. Since any finite axiomatization over A2 has the

form O(V), this shows that A* is essentially not finitely axiomatizable
over A2.
In Ackermann's original paper introducing the theory A, he also dis-

cussed an extension Aoo of A. Aoo is based on the idea that there can be
sets of different "orders", so that something which appears as a class
from the poi:at of view o f "order o:ae" may be a set of "order two".
Thus A ~ has individual constants V 0, V l , .,,, ; the old sets (members of
V) become the sets of order zero (members of V0), and V l , V 2, ... cor-
respond to the sets of higher orders. Also a final class Voo is introduced
which includes all the lower orders. We give here Levy's form ot Ao,,
(see Levy [8i ), which is slightly stronger than Ackermann's originat
version.
Definition 4.3. a. A,,~ is the theory (formulated in a language with

binary predicate e, and individual constants V, (n ~ ~ ) and V ~ ) whose
axioms are the universal closures of A1 (er.tensionality),
A2' (class construction schema)
Vz At(t ~t z ~ , t ~ V=,^ O)
(where z does~iaot occur in O)

4t
A31~ (n _< ~, - strong completeness for V n) Sc(V,),
A4'n (n _< ~ - modified Ackermann schema),
Section 4 233
x, y ~ V,, ^ A t ( ~ ( t . x. y ) ~ t ~ V,,)
Vz e V At(t e z ~ ¢ ( t , x . v'D
where ¢ is a formula involving only V 111 f o r m <~ n, a n d z d o e s n o t

o c c u r in ~,
A6' Vn C
- V,~ +I c- Voo (fornEco).
b. A * is obtained from A ~ by adding the regularity axioms
A5' A x c V [Vtt(u~x)-- Vu(u6xÂt(t~x~tq~u))]
for each n < ~ .
A simpler characterization of A ~ is the follow ing.
T h e o r e m 4.4. Aoo m a y be a x i o m a t i z e d by
A ( V , ) all sentences ~(V,,) such that ¢ ( V ) is an a x i o m o f A (where
A6 Vn ~ V +1 EVoo (wherenEco).
Proof. Evidently A,,o yields AI(V,,)= AI, A3(V,:), A4(V,z), and (using
A6') A 2 ( V , ) for each n < o~. To see that A6 holds, in A4' take t ~ V m
for ~ and m + 1 for n. and conclude (using V ~ ~ V m + 1 ) that
V m E V m + j . By A3'. V m+l E V ~ also, so A-3 holds.
In the o t h e r direction, A I , A2', ~3' are assumed, and A6' follows
t ,
from A6 using A3'. Now it is easy to see that A4 ( V , ) yields A4 n ,
t
A6 assures that the parameters Vi, i E n may occur in the ¢ of A4 n .
T h e o r e m 4.5. l d e n t i f v V 0 w i t h V. Then .for each e, V-sentence o,

A* 1- o iff A * I--- o.
P r o o f . E v i d e n t l y A* b o i m p l i e s A * I- o. Let ¢ be a f o r m u l a w i t h one
free variable x, and suppose that ~(V) is provable in A * . Then by the
preceding t h e o r e m , there is a formula O with one free variable x such
that A * b O ( V ) and
I-- O(V o) ^...^ O ( V ) ^ O ( V ~ ) ^ V o ~ ... E V n e Voo -~ ~0(Vo).

234 A ckermaml's set tk,,ory equals ZF
Thus ~t will suffice to prove in ~ * the existential closures o f the for-

mulas
1) I ~ ( V ) A{~(Vn) A...A O ( V 0 ) A V e O,1 E ... E O0
(where A* I-- O(V), and n ~ co). ( N o w o 0 plays the role o f V~o, % that
o f V 1.) The p r o o f is by i n d u c t i o n on n. Assume the result for m . and
let "4, be the existential closure o f 1) (taking n = m). Then the induction
hypothesis yields
A* I-- q,(V)
and by til~ upward reflection principle ( 1.7d) we get
2) A* I- VV,n+l [V C ore+ , A q,(v,n ÷l )1 "
Evidently 2) implies tile existential closure of 1 ) for tl = m + 1. This

completes the proof.
4.2. N a t u r a l m o d e l s o f A.
We now proceed to some rer.aarks on natural models o f A.
Definition 4.6. Let o~,/3 be ordinals, n ~ a ~/3. We write Ra -< R¢3 iff for
It
every 3' ~ Ro~,
(Ro~ + n, e, x> --- (R/3 + n , e, x , ' .
Remark. Ra -< R/3 iff Ra <: R/3, and Ra -< R/3 means that R/3 is a 2nd
0 1
order extension o f Ro~. We could as well write
(Ra + n, e, x ) x E Ro~ =- (14,[3 + tl, e , X ) x E R a '
(using a c o m m o n n o t a t i o n ) because finite sequences are representable

in these structures.
The condition R, < R@and.0 E 0 is fairly strong, as will be shown in
1
Theorem 4.12, The next theorem shows that the condition is weaker
than certain conditions which have been discussed in the literature.
Definition 4.7. Let /3 be an ordinal, and n E p. We say p is n-indescribable

iff for every first order sentence u involving e and one additional unary
predicate symbol X, and every X C_Rp, if (Rp + n, E, X> l= 0 then there
isanordinalcxE@suchthat(Ror+n,e,Xn Ra>l=u.
For rt E Y, we say that ,8 is izth order indescribable iff for every
t)t E II, 0 &s m-indescribable.
Remark. It is known for example that if K is a measurable cardinal then

there is 0 E K such that for every n E p, 0 is n-indescribable (Vaught
[ 20 I; also see Reinhardt-Silver [ 141 and Silver [ 161, ch. 4). The notion
of indescribability was introduced by Hanf and Scott in 171. It should
be remarked that p is 2nd order indescribable iff (R,IJ + 1, E) is a model
of Fernays strong set theory (see Bernays I2 I ). 0 is strongly inaccessible
iff 0 is O-indescribable.
Theorem 4.8. If‘p is (n + i )-ittdescribable, then thew ;-

such slutt
Sketch of proof. In the definition of (n + 1)-indescriballlrity take
X = (<y, 6) : (RP + n, E) t= O[y] and y E Rp) .
(Evidently 0 = kI(3 so X C_RB). There is an e-formula 8 with one free

variable such that for every ordinal /3, every e-formula 8, and any
YE R,
,6 Ackermann's set theory equals ZF
Now apply indescribability: since we have
(R/t + n + 1, e, X ) l= A t ~ R/~(t ~ X - ~ ; xI,(t))
(of course R# is expressible in terms of X and e), we may choose c~.c

such that
(Ra+n+l,e, XnRc:)b AtER~(tEX< ,,I,(t)).
Consequently, for all y E R a , and all e-tbrmulas 0,
(y,O) E Xn Ra iff ( R a + n , e ) p 0 [ y ]
iff (R~+n,e)~0[yl ,
and thus Rt~ < R#.

n
Theorem 4.9. f f Rt~ < R/~ a n d n E ~ E ~, t h e n

n
(R/~ + n, e, R a ) ~ A .
Proof. It is enough to show that (treating t~ as a constant) the schema
1) xE RaÔ(x,a)-* Vy~aO(x,y)
holds in (R~ +n, e,a), i.e. in (P,fl +n, e) w h e n ~ denotes a.

First observe that since (Ra, e) -< (RtLe), we certainly have
x, y E a implies x + y E ot
(where + is ordinal addition).

Let A(o o, v l) be a formula expressing
'Oo,O1 are ordinals, and there is an ordinal whose order
type is the ordinal sum vo + 01'.
Section 4 237
Evidently for all ordinals ~, n, if n ~ ~ and x ~ R~ + ~, we have
2) ( R ~ + n , e ) l= A [ x , n ] iff x e ~ .
Now the hypothesis of the theorem is that for x E R~,
<R~ + n, e,x> = (Ra + n, e , x > ,
which is to say that (treating a, n as self-denoting constants) the schema
3) x o ..... Xm_1 e R~ -~ [~< > CR~+"I
holds in <Rt3 + n, e> (here Xo, ..., x m_) exhaust the free variables of ~o).
Evidently (by 2))
4) x ~ R a ^ O(x, a ) - , Vy[A(y, n)A O ( x , y ) I
holds in <R~ + n, e>, so app;ying 3) (witb parameters x, n ~ Ra) to 4)

we get
-+ V.): E R o t + n l A Rot+n (y, n) ^ O Ra+n (x, y)]
so 2) yields
Vy e Rt~[® Ra÷n (x, Y)]
and thus 3) yields the schema 1) as desired.
Remark. This shows (assuming the appropAate indescribable cardinals

exist) that <R~, e, Rat I= A does not imply <Ra, e> -< <Rt3, e), and even
that <R~, e, Re> ~ A and <R#, e) ~ Z does not imply (R~, e> ~- ZF. For
suppose that Ra < R/~; then <Rt~ + 1, e, RcC ~ A, but evidently not
1
(Ra, e> -< (R/3 + 1, e>. Or suppose that Ra -< R/3; then <RE + ~ , e, Ra>
tO
A and <R# + co, e) ~ Z, but evidently not (P/3 + ~ , e) ~: ZF.

238 Ackerm~m'~ set theory equals ZF
We will now show (T~eorem 4.12) a weakened converse of Theorem

4.8. Theorem 4.12 is known (the proof for the c~se n = ! is carried out
in Tharp [24] ; the general statement is given in Reinhardt [ 15a] ); for
completeness we include the proof here. First a definition.
Definition 4.10. Let O be an e-formula with free variables x, y. Let # be

an ordinal, and A c_ R/3. We say that O well orders A in R/3 iff
{(u, o> : (Rg, e) P O[u, o] and u, o ~ A}
is a well ordering of A.
It is well known that the axiom of constructibility provides nice well

orderings. In particular we will need
Theorem 4.11. (G6del). S u p p o s e that V = L. Then f o r every limit

ordinal ~ >_ o~, there is an e-formula 0 such that 0 wel! orders R~ + 1 in
R~+l.
For proof see G6del [ 5 ].
Theorem 4.12. S u p p o s e that 04= n ~ ~, Ra -< RO, and that ¢ i s a f o r -

mula which well orders R# + 1 in Rf3 + n. Tlten ~ and ~ are n-indescrib-
able.
Proof. To simplify the discussion we assume n = 1 ; all the ideas are con-
tained in this case.
It is enough to see that for every e-formula O (with one free variable)
and every X c R~, we have
1) i f ( R # + 1, e) ~ ®[X] then there exists a ~ # such that

(R~ + 1, e) ~ O[X n Rc~].
Suppose that 1) fails. Then there is an e-formula ® (with free variable x)

and an X c_ R# such that
2) O(x)^~ V o ~ 5[0R°~+l(x n Ro~)]

Section 4 239
holds in (Re + 1, e> w h e n x takes the value X. (Here we use the defina-
bility oft3 in (P,43 + 1, e); in the general case e n e can use the definability
o f # in (Rt3 + n, e, n>.) Let ",I,( x ) b e the formula displayed in 2). Then
since ~, well orders Re + 1 in R/3 + 1,
3) V! x('4, ( x ) ^ Ay[~0(y, x) ^ q' O,') --~ x : y] )
holds in ( R/3 + 1, e), and hence (since Re -< R/3) in <Ra + 1, e> ,also. Let
1
X' be the subset o f Re defined by 3) in (Re + 1, e), and X be the subset
of R/3 defined by 3) in ( R~ + 1, e). Since Re -< R/3, for each x ~ Ra we
1
have
x ~ X' iff x ~ X .
Thus X' = X c, Re. By the definition o f X', it must satisfy xo (and

hence 0) in (R,~ + 1, e), so we have
4 ¸) (R/3 + 1, t) ~ Ve ~/3 ®Re+l [ X ' I .
Thus since X' = X n Re, X must satisfy
Ve ~/3 0 Re+l(x C~ Re)
in (R~ + 1, e). which contradicts (R/3 + 1, e) N q, [ X ] . Thus/3 must be

1-indescribable. Since for each formula O, and every ordinal/3, the con-
dition
'for every X c_ R/3, 1) holds'
is expressible in <R/3 + 1, e>, we see that a is also 1-indescribable. This

c o m p l e t e s the proof.
Remark. T h e o r e m 4.12 shows that if V = L, and Re -< R/3, e ~/3, then

t
e is greater than the first inaccessible cardinal, thus settling a q u e s t i o n
o f Levy and Vaught in case V = L (the general question remains open).
4.3. Questions related to A about definability in natural models.

We now raise some questions about definability in natural model~
which are related to Ackermann's set theory.
Theorem 4.9 generalizes a result of L~vy's (he obtained the case
n = I ). Is there a second order version o f Grewe's theorem 2.8?
Question 4.13. Let o~, fl be ordinals. Suppose that for every x ~ Ra, a is
not definable in (R~3 + 1, e, x). Is there a 7 <- # such that a ~ ~, and
Ro~ -< R3,?
l
Theorem 4.9 shows (assuming there are various indescribable cardi-
nals) that we can have a wide variety o f ordinals a, # such that
(R#, e, R~) ~ A but not Ra --- R~. The methods of proof do not seem
to answer the following question, however.
Question 4.14. Suppose that a,/3, ~, are ordinals,/3 ~ 7 and R/3 -< RT,
and for every x ~ Ra, o~ is not definable in ( R/3, e, x). Does it follow
that Ra -< R/3?
Remark. This is equivalent to the question whether we must have

Ra -< R/3 if (RT, e, Ra), (RT, e, R G) are models of A with ~ <_/3. The
syntactic question corresponding to this is whether the schema
1) x ~ V 0 - , [O v o ~ e v~]
(where x is the only free variable of the e-formula 0) is probable in A * .

This question is related to Theorem 3.c~~, which says that (in A*) the
quantifier 'Vx ~ V' can be eliminated in favor of an expression (which
has a parameter y ~ V) involv!ng only e. Improving Theorem 3.9 to
read:
2) For every e-formula O with one free variable x there is an e-formula

e) with one free variable x such that
A I - x ~_v-* [O v ~ ~ 0 ]
would immediately yield 1) and thus a positive answer to Question 4.14.

Section 4 241
The following question is more closely related to type theory with

typical ambiguity (see Specker [ 18] ) than to A, but seems to have the
flavor of the preceding questions.
Question 4.15. How large is the first ordir.al 3' such that there are
ordinals a i, i ~ ~ for which i < / - + ai < e / a n d
3) (R3', e, Oti)i-to =- (R')', e, 0ti+l)iEto 9
Remark. If 3' satisfies the condition ~, -+ ( ~ ) < ~ (introduced by Silver

$V
[ 161, p. 78), then it is easy to see that 7 satisfies 3). (Silver's condition
is satisfied if 3' is a measvrable cardinal.) But the condition 3) is prob-
ably quite weak.
4.4. Q u e s t i o n s a b o u t f i n i t e axio.matizability.
Next we consider some questions about ffinite axiomatizability.
In most cases where we have a proof that a theory T is not finitely
axiomatizable over a theory S, the methods of proof yield a stronger
result, namely that if S c_ T' c_ T and T' is finitely axiomatizable over S,
then the consistency of T' is provable in T. (Of course a precise formu-
lation of the stronger result is sensitive to the way ' I " is consistent' is
expressed.) I would like to mention some examples of theories T, S
such that the consistency of S is not provable in I-, but it appears likely
that T is not finitely axiomatizable over S.
Conjecture 4.16.
a. A + Z is not finitely axiomatizable over A.
b. A + ZF is not finitely axi~matizable over A.
The weak comprehension schema of' Z - was introduced and studied
by Tarski. He showed that Z is essentially not finitely axiomatizable
over Z - , and that Z- is finitely axiomatizable (unpublished). I do not
even know whether A + Z- I-- Z. Of course b implies a. We can make
some conjectures even stronger than b. For this we need some defini-
tions.
Definition 4.17. The theory ZA is formulated in the language of A,

with axioms A1, A2, A3, A5 and the following schema of complete
242 Ackermann "sset theory equals ZF
reflection:
x,y~V~[ov ,--~O]
(where ® is an e-formula whose only free variables are x, y).
The theory ZA was introduced by Vaught (as a technical device) in

connection with Ldvy's p r o o f of the consistency of A (see [ 8 ] ). For
completeness we mention the main result relating ZA and ZF.
Theorem 4.18.
a. F o r e-selztences o, ZA I-- o i l l Z F t- o.
b. Ac_ZA.
Sketch of proof. See Levy [8]. a. ZF ~ ZA is essentially Theorem 2.7

of Levy, Montague, Vaught. The proof that for e-sentences ZA I-- o im-
plies ZF I---o uses the reflection principle for ZF (Montague [ 12] ).
b. It is enough to check DF c_ ZA, as the reflection principle for V
(Theorem 1.7c) is contained in DF + ZF and implies Ackermann's
schema A4. (DF is defined by Definition 3.3.) Notice that a and b to-
gether show Levy's half of t h e o r e m 3.1.
Definition 4.19. Let T, S be theories in a language with e, V-formuias.

We say T is neatly interpretable in S iff there is an e-formula ,p with one
free variable x such that
a. T I--- o i m p l i e s S I-- o ~' ,

b. S ~- (x E y v x -C- y ) A ¢ ( y ) -~ ~p(>.),
c. S I - y C V--, ~0(y).
Remark. Grewe showed in [61 that A + Z- is neatly interpretable in A*
Conjecture 4.20.
ZA is not finitely axiomatizable over A + DF.
Question 4.21. Let T be neatly interpretable in A + DF. Does it follow

that Z is essentially not finitely axiomatizable over T?
Section 4 243
A positive answer to 4.21 implies 4.20 which in turn implies 4.16b.
The last example seems in some sense more elementary. The theories
have infinite similarity type, but this defect may be removed by simply
considering the e-part of the theories.
Question 4.22. T will be a theory formulated in a language w i t h e and

individual constants V n (for each n ~ ~ ) . The axioms of T include (for
each n) the pairing, union, and power set axioms relativized to V n , the
schema A2 with V replaced by V n , and an axiom V n ~_ V n t 1. Is T + Z
finitely axiomatizable over T?
4.5. C o m p a r i n g A a n d A*. Since A* is interpretable in A (by taking Ro,

for V), and A* is firAtely axiomatizable over A by definition, the theo-
ries are of essentially the same strength (although it is easy to prove
that A* 4: A). Moreover the two theories have the same natural models.
Thus, both from the point ofvizw of relative consistency, and from the
point of view of natural models, questions about what is provable in A
without use of A5 are curiosities. For completeness, however, I mention
the following questions which 1 have not answered.
Question 4.24. a. Is the replacement schema relativized to V provable

in A?
b. A ~ V z A t ( t ~ z ~ t =V)?
C. If we add the reflection principle for V (1.7c) to A do we obtain
A* ?
244 Ackermann "s set theory equals ZF
Section 5
This section has two parts. The first part extends the main result o f
Section 3 to certain extensions of ZF and A. The second contains an
analysis of the intuitive ideas behind Ackermann's theory. This ana2ysis
suggests the introduction of an extension A + of A.
5. I. Extension o f 3.1. To illustrate the method we extend Theorem 3.1

to ZM and A~.
Definition 5.1. a. ZM is the set of e-sentences provable in tile theory

obtained from ZA (see Definition 4.17) by adding rhe axiom
e0 x~VAAu~x Vo~VRl(u,o,r)-+
VyEVAu~x Vo~y Rl(u,v,r)
b. A~ is obtained from A* by adding % .
Remark. Notice that e 0 is a replacement axiom; it assures that a is in-

accessible. ZM and stronger theories ZM A have been studied in the
literature. (See especially Levy [8, 9]. ZA is the theory Sb of [81 .)
Other characterizations of ZM are known; the one we have chosen for
5. la is due to Levy and will be convenient in the proof of 5.2. Results
analogous to 5.2 hold for ZM A , and indeed can be proved in the same
way.
Theorem 5.2.
A~l--o v iff Z M | - o .
Proof. First we show that if ¢ is an e-formula with free variable x, and

~(V) is any theorem of ZA, then
1) A* ~ VV'(~0V'(v) A V ~ V' A Sc(V')).
It is enough to show that for any finite set F of axioms of ZA, we can
show
Section 5 245
2) A* t-- VV'(V ~ V'^ Sc(V') a /~ oV').

o~F
Now the axioms of ZA are A1, instances of the schema A2, A3, and in-
stances of the complete reflection schema
3) Ax E V(Ov , . , O)
(where O is an e-formula with one free variable x). (Arguil g informally

in A* now) for an7 V' such that V c- V' and Sc(V'), it is evident that all
instances of A2 hold relativized to v ' , and the relativizatiens to V' of
3) become
4) A x E V(® v' ,--~ ®v').
But now Corollary 3.11 allows us tc :hoose V' so that V E V' and
Sc(V') and any finite number of instances of 4) hold. This establishes
2) and hence 1).
Now evidently A~ c ZA + e 0, so if A~ l- ov then ZM I-- o. If
ZM I- o, then ZA I- e 0 -~ ov . By 1), we then have
A* t- VV'(V ~ V ' ^ S c ( V ' ) n ( e 0 ~ o v ) v ' ) ,
and since
A*t-V~V'ASc(V')-~[%-~e ]^[(ov) v ~o vl ,
it follow, that A* ~- e o ~, ov , i.e. A~ ~- ov as desired.
5.2. On "'sharply delimited collections". The next paragraph is devoted

to an analysis o f the intuitive ideas behind Ackermann's theory.
It was mentioned in the introduction that the basis for Ackermann's
axioma schema A4 appears to be the idea that sufficiently sharply de-
limited collections of sets are sets. In this connection I would like to
suggest that it is helpful to think of " t h e collection of all sets (V) is not
sharply delimited" as meaning that there are various classes which are
possible interpretations of the collection of all sets, among which there
246 Ackermann "sset theory equals ZF
is n o clear choice. Thus the individual c o n s t a n t V is t o be t h o u g h t o f as

a variable which could d e n o t e any o f (say) V 0' V l , V 2 , .... (This sug-
gestion seems to be c o n s o n a n t with the theory Aoo considered by
A c k e r m a n n . ) In order to formulate the principle 'sharply d e l i m i t e d
collection o f sets are sets' we m u s t explain 'sharply d e l h n i t e d ' also for
(at least some) collections o t h e r than the collection o f all sets. If ~(t) is
any formula (with one free variable t) we explain the n o t i o n for the col-
lection o f sets satisfying ~0(t). This collection corresponds to the formula
,I,(V, t) = ~0(t) ^ t ~ V; thus it is natural to say that the collection is
sharply delimited if w h a t satisfies ,!, (V, t) is i n d e p e n d e n t o f the inter-
pretation o f V. In order to simplify the discussion let us assume there
are exactly t w o interpretations o f the collection o f all sets, V 0 and V l ,
and that V 0 c_ V 1" Now the formula
1) A (q,(V 0 , t ) , , ,J,,(V 1 , t))

t
expresses that the question w h e t h e r t satisfies q, (V, t) is i n d e p e n d e n t o f

the interpretation o f V, i.e. { t : ,J.,(V, t) } is sharply delimited. T h e n the
principle'that sharply delimited collections o f sets are sets is expressed
by
2) A (q,(V 0, t ) , , q,(V 1 , t)) ~ Vz ~ V 0 A t ( t ~ z ~ q,(V 0, t ) ) .

t
Evidently if ~(t) does n o t involve V, and we have
3) At(~o(t) ~ t ~ V 0)
then
4) At(:(t)^ t~ V o , ,~, ~ ( t ) ^ t ~ Vl)
so that we have by 2)
K t ( ¢ ( t ) ~ t ~ V o) ~ Vz ~ V o A t ( t ~ z ~ ~ ~o(t)),
which, if we identify V 0 and V l , is a special case o f A4. We get A4 itself

Section 5 247
simply by allowing x, y as free variables in ®, adding 'x, y ~ V 0' to the

h y p o t h e s e s o f 2).
These considerations suggest the following definition.
Definition 5.3. A + is a theory (in a language with binary predicate e and

t w o individual constants V, V') whose axioms are
1. The axioms o f A* e x c e p t for A c k e r m a n n ' s schema A4.
2. VC_ V'.
3. The universal closures o f the following schema (in place o f A4)
A4 ÷ x,y~ V ^ A (~(V, x , y , t) A t ~ V ~:-" ~(V', x , y , t) A t ~ "V')

t
Vz ~ V At(t ~ z *--~ t ~ V/x ¢(V, x , Y, t))
(where ~ is an e-formula whose free variables are a m o n g v 0, v 1 , 0 2 , 1)3

and z does not occur in ~).
Remark. It is not hard to show that A + is consistent if A* plus the fol-

lowing schema o f indescribability is consistent:
i) x, yC_ V ^ @(V,x,y)--~ V v ~ _ V O ( v , x N v , y N o ) ,
where 63 is any e-formula with free variables v 0, v l, v2 ; o f course n is to

be an appropriate representation o f intersection. For an indication of a
p r o o f see [15a; Chapter 5, p. 7 8 ] . The schema 1) was considered by
Vaught [20] and shown to be consistent with A* if ZF + ' t h e r e is a
measurable cardinal' is consistent. For a stronger result, see Reinhardt-
Silver [ 14] ; prools can be f o u n d in Silver [ 16]. It seems likely that A*
plus l ) is m u c h stronger than A + , but l k n o w no resultz, in this direc-
tion.
T h e o r e m 5.4.
a. A * ~ A + ,
b. A + t- V ~ V ' .
Proof. Evidently we have already proved a. To prove b, we argue infor-

mally in A + . Let ~ range over ordinals, and V = Ra. If
248 Aekermann's set theoo' equals ZF
Ro~={x" V#~V(x~R~C- V)}=

={x" Vfl~V'(xER/3CV')},
thc~n b y A 4 + , R a ~ R a , w h i c h is impossible. T h u s w e m a y c h o o s e 13 ~ V'

so t h a t RO~ V' a n d R/3 _~" R e . N o w a =/~ v ~ ~ a -~ R/~ C R a , so f o r this
/~, a ~ ~ a n d we h a v e V = R a ~ R/3 c_ V' as desired.
References
[ 1] W.Ackermann, Zur Axiomatik der Mengenlehre, Mathematische Annalen 131 (1956)

336- 345.
12] P.Bernays, Zur Frage der Unendlichkeitsschemata in der axiomatischen Mengenlehre,
Essays on the foundations of mathematics (Magnes Press, Jerusalem, 1961) 3-49.
[3] G.Cantor, Beitr~ige zur Begriindung der transfiniten Mengenlehre, Gesammelte A bhand-
kungen (Springer, Berlin, 1932) 282.
14] A.A.Fraenkel and Y.Bar-HiUel, Foundations of set theory (North-Holland, Amsterdam,
1958).
[5] K.G6del, The consistency of tile axiom of choice and of the generalized continuuva
hypothesis with the axioms of set theory, fovrth printing (Princeton, 1958) 69 Pl •
[6] R.Grewe, On Ackermann's set theory, Doctoral dissertation, University of California, Los
Angeles, 1966.
[6al R.Grewe, Natural models of Ackermann's set theory. J. Symb. Logic 34 (1969) 48 ! ....
488.
[7] W.Hanf and D.Scott, Classifying inacceesible cardinals, Am. Math. Soc. Notices 12 (1965)
723.
[8] A.Levy, On Ackermann's set theory, J. Symb. Logic 24 (1959) 154-166.
[9] A.Levy, Axiom schemata of strong infinity in axiomatic set theory, Pacific J. Math. lf~
(1960) 223-238.
[ 101 A.Levy, A hierarchy of formulas it, set theory, Memoir Number 57, American Mathe-
matical Society (Providence, 1965).
[ 11 ] A.Levy and R.L.Vaught, Principles of partial reflection in the set theories of Zermelo and
Ackermann, Pacific J. Math. 11 (196t) 1045-1062.
[ 12] R.Montague, Two contributions to the foundations of set theory, Logic, Methodology
and Philosophy of Science. Proceedings of the ! 960 International Congress (Stanford,
1962) 94-110.
[ 13] R.Montague and R.L.Vaught, Natural models of set theory, Fund. Math. 47 (1959) 219-
242.
[ 14] W.N.Reinhardt and J.Sibmr. !. On some problems of Erdos and ltajnal, Am. Math. Soc.
Notices 12 (1965) 723.
[151 W.N.Reinhardt, Ackermann's set theory coincides with ZF, Am. Math. Soc. Notices 13
(1966) 727.
[ 15al W.N.Reinhardt, Topics in the metamathematics of set theory, Doctoral dissertation,
University of California, Berkeley, 1967.
References 249
[ 16] J.Silver, Some applications of model theory in set theory, Doctoral dissertation, Univer-
sity of California, Berkeley, 1966.
[ 17] R.Solovay. Real-v~ued measurable cardinals, Am. Math. Soc. Notices.
[ 18] E.Specker, Typical ambiguity, Logic, Methodology and Philosophy of Science. Proceed-
ings of the 1960 International Congress (Stanford, 1962) pp. 116-124.
[ 19] A.Tarski and R.L.Vaught, Arithmetical extensions of relational systems, Compositio
Math. 13 (1957)
[ 20] R.L.Vaught, Inde~ribable cardinals, Am. Math. Soc. Notices 10 (1963) 126.
[ 21] E.Zermelo, Untersuchungen tiber die Grundlagen der Mengenlehre I., Mathematische
Annalen 65 (1907) 261-281.
[22] G.Kreisel, Mathematical logic, Lectures on Modern Mathematics, vol III, ed. E.Saaty
(Wiley, 1965).
[ 23] H.J.Keisler and A.Tarski, From accessible to inaccessible cardinals, Fund. Math. 53 (1964)
225--308.
[ 24] L.Tharp, On a set theory of Bernays, J. Symb. Logic 32 (1967) 319-321.

Ackermann's Set Theory

Uploaded by

Copyright:

Available Formats

Ackermann's Set Theory

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ackermann's Set Theory

Uploaded by

Copyright:

Available Formats

ANNALS Oi: MATHEMATICAL LOGIC -~ Volume 2, No. 2 (1970) pp.

ACKERMANN's SET THEORY EQUALS Z F *

Received 26 Janu,'~y 1970

Section 0.1 introduces Ackermann's set theory and contains a brief

O. 1. A ckermann's set theor.v

At(t~x ~ t ~j,)~ x =y.

A2. (Class construction)

where O is any formula in ~ and V not involving z.

xE V^ [tEx v Att(uE t ~ uEx)] ~ tE V.

A4. (Ackermann's schema)

x,y EVA At(O(x,y, t)* 1E V

We will let A be the theory determined by these axioms. A* is o1: ~ained

A5. (Regularity for sets)

Perhaps some remarks on the motivation for the theory A will be

(1) In the paper introducil.g the theory A [1 ], Ackermann showed that

(1) If the relativization to V of a sentence o is provable in A*, then

(2) If o is a theorem of ZF, then the relativization to V of o is provable

of regularity for sets. Since A* is consistent relative to A in much the

(,2) A * I-- , I , ( V ) --, V ' v ' ( q , ( V " ~ A V ~ V' A

•This is stronger than (1). Sections 1 and 2 provide necessary prelimi-

oF A* only in that it has some new individual constants. (III) If (R/3,~ e)

0.3. Set theoretic notation

junction Y4 ,1, m . If u is a variable o r individual c o n s t a n t , a n d v is a

o b t a i n e d f r o m 0 by relativizing all quantifiers to ~ (A uO goes to

w h e r e o' is the first variable distinct f r o m o not o c c u r r i n g in 69.

where 0 is any f o r m u l a in w h i c h z does n o t occur. Zermelo-FraenLel set

where z does not o c c u r ;,n ~p. Q is the axiom

Z + is the t h e o r y Z + Q ( R n ( y , r) is a f o r m a l i z a t i e n o f r = Ry; for an ex-

( w h e r e z ~ y and z does not o c c u r i~: O). Z 0 is Z w i t h o u t f o u n d a t i o n .

0.5. Semantical notation

(See Tarski-Vaught [ 19] for discussion o f these semantical notions;

In this Section we define a number of e-formulas we will have need

1.1. S o m e e - j o r m u l a s . We will have need to refer to the following e-for-

Definition 1.1. a. x g y is the formula A t ( t ~ x -* t ~ y ) and is read

(Ay, z)(z E y A.v E x ~ z E x)

'x is complete (or transitive)'.

'x is supercomplete (or strongly transitive)'.

Cp(x) A (Ay, z, w 6 x)iy ~ y A ( y E Z, V F = Z V Z E ) ' ) A

/~(w~y~z~w~Z)IA Au[u ~ x ^ Vt(t ~ u)-*

Vt(t E u A ~ V s ( s ~ t ,^ s ~ tt))] /',

'z is the doubleton { x, y } '

'z is the ordered couple (x, y ) ' .

'x is in the relation z to y ' .

Ord(x) A Vz R n k ( x , y, z), where

AAu, o{[uE.x'ARI(u,v,z)] vlu =xAv =y]

-~ A t [ t e v *--~ V u ' , v'(u' ~ u A RI(u', v', Z)A t C_ v')] }

Rn(x, y ) is read 'y is the sets o f rank less than x'.

'.,," is a limit ordinal'.

1.2. Absoluteness. We will occa-sionally need to observe t h a t certain e-

Definition 1.2. Let ~f be a t h e o r y and let U ~-e a formula o f the language

w h e r e the free variables of 0 are e.mong v 0 . . . . . On-l"

If V is an individual constant, then instead o f saying absolute with re-

(Let us refer to this axiom as Sc U.)

Lemma 1.3. L e t T be a theory, U a f o r m u l a o f the language o f T whose

T I- Uo 0 A..A Urn_ 1 A Vx O ~ Vx(Ux A O ) ,

where the free variables o f O are a m o n g v o .... , % - 1 , then V x 0 is abso-

Then the set o f f o r m u l a s I" absolute with respect to U over T contains

The proof of Lemma 1.3 is trivial; part b follows from part a.

Using 1.3 and the trivial facts

one easily checks that