This preview shows page 1. Sign up to view the full content.
Unformatted text preview: AXIOMATIZATION OF SET THEORY BY EXTENSIONALITY, SEPARATION,
AND REDUCIBILITY
SEMINAR NOTES
by
Harvey M. Friedman
Department of Mathematics
Ohio State University
Columbus, Ohio 43210
friedman@math.ohiostate.edu
www.math.ohiostate.edu/~friedman/
October 5, 1997
We discuss several axiomatizations of set theory in first
order predicate calculus with epsilon and a constant symbol
W, starting with the simple system K(W) which has a strong
equivalence with ZF without Foundation. The other systems
correspond to various extensions of ZF by certain large
cardinal hypotheses. These axiomatizations are unusually
simple and uncluttered, and are highly suggestive of
underlying philosophical principles that generate higher set
theory.
The general philosophy is that at any time, one can attempt
to conceive of the entire set theoretic universe. But
according to Russell's Paradox, as one later conceives of the
entire set theoretic universe, reflecting on the previous
conception, one obtains a yet larger set theoretic universe.
However, according to this general philosophy, nothing really
new is obtained  the first set theoretic universe is, for
all intents and purposes, the same as the second set
theoretic universe, even though it is not literally the same.
This principle takes the form of a Reducibility axiom that
can be viewed as an extension of Russell's Reducibility.
We view the variables in these axiomatizations as ranging
over today's set theoretic world, and the constant symbol W
as representing yesterday's set theoretic world (the subworld). In K(W), it can be proved that W is transitive  no
element of yesterday's world picks up any new elements from
today's world. In some of the stronger systems which generate
large cardinals, W is provably non transitive  some elements
of yesterday's world do pick up new elements from today's
world.
This is supported by a natural weakening of separation, with
a corresponding strengthening of Reducibility, which allows 2
for the construction of a standard model of ZF + measurable
cardinals and beyond. W gets interpreted as images of
elementary embeddings associated with large cardinals.
The systems K(J),K*(J) give direct axiomatizations of an
elementary embedding J, and are much more committal than
K1(W)K5(W). They postulate a “projection function” from
today’s world into (onto) yesterday’s world, which is related
to published ideas of Reinhardt. We include them because of
their special simplicity.
Here is the organization of the results.
1. ZF\Foundation.
Extensionality (EXT), Subworld Separation (SS),
Reducibility (RED).
2. Indescribable and subtle cardinals.
EXT, SS, RED+.
4. Measurable cardinals.
EXT, SS, RED, TRANS.
5. Elementary embeddings from V(a) into V(b).
EXT, SS, RED, RED’, TRANS.
6. ZF\Foundation again.
EXT, SS, RED, RED’’.
7. Elementary embeddings from V(a) into V(a).
EXT, SS, RED, RED’’, TRANS, Largeness.
8. Elementary embeddings from V into V.
EXT, SS, RED, RED’’, TRANS, Subworld Collection.
NOTE: SS < SS; RED+ > RED; and RED’’> RED’.
Largeness says that every element of W can be mapped oneone
into the intersection of W with an element of W. TRANS says
that every transitive subset of W is a subset of a fixed
element of W.
9. Axiomatic elementary embeddings: huge cardinals and V(l)
into V(l).
EXT, Separation, Elementarity, Fixed Points.
10. Axiomatic elementary embeddings: V into V.
EXT, Replacement, Elementarity, Initial Fixed Points.
In each case, we prove a close relationship between these new
systems and standard formal set theories without Choice,
sometimes without Foundation. Foundation is inessential here 3
in that the cumulative hierarchy provides a standard
interpretation.
The absence of the axiom of Choice is a more delicate matter.
Until we move significantly past the measurable cardinal
level, we can use known inner model techniques going back to
Gödel ablish the Œ interpretability of the axiom of Choice.
At the higher levels, in some cases we can invoke some
unpublished work of Hugh Woodin that uses forcing models
instead of inner models to establish Œ interpretability of
the axiom of Choice  at the cost of weakening somewhat the
large cardinal axioms in the forcing model. So, more or less,
the absence of the axiom of Choice does not substantially
weaken higher set theory with respect to Œ interpretability.
Also, of course, there is nothing to prevent you from simply
adding the axiom of choice in any of its myriad forms to our
systems. Then the standard set theories that they correspond
to will simply be standard set theories with the axiom of
Choice. We choose not to do this, since in doing so, there is
no interaction with the rest of the axioms that we use.
We begin with a discussion of the system K(W) based on
L(Œ,W), which is first order predicate calculus with the
binary relation symbol Œ and the constant symbol W. The
axioms of K(W) are:
1. Extensionality (EXT). ("x1)(x1 Œ x2 ´ x1 Œ x3) Æ ("x1)(x2
Œ x1 ´ x3 Œ x1).
2. Subworld Separation (SS). x1 Œ W Æ ($x2 Œ W)("x3)(x3 Œ x2
´ (x3 Œ x1 & j)), where j is a formula in L(Œ,W) in which x2
is not free.
3. Reducibility (RED). (x1,…,xn Œ W Ÿ j) Æ ($xn+1 Œ W)(j),
where n ≥ 0 and j is a formula in L(Œ) whose free variables
are among x1,…,xn+1.
Define x = y iff ("z)(z Œ x ´ z Œ y). Using EXT:
LEMMA 1.1. Let j be a formula in L(Œ,W) without x2. K(W)
proves: x1 = x1. x1 = x2 Æ x2 = x1. (x1 = x2 & x2 = x3) Æ x1 =
x3. x1 = x2 Æ (j ´ j[x1/x2]).
For any formula j in L(Œ), let jW be the result of
relativizing all quantifiers to W. 4
LEMMA 1.2. Let j be a formula in L(Œ) whose free variables
are among x1,…,xn, n ≥ 0. The following is provable in K(W).
x1,…,xn Œ W Æ (j ´ jW).
LEMMA 1.3. The following is provable in K(W). ($x1)(x1 Œ W).
x1 Œ x2 Œ W Æ x1 Œ W.
The idea is to form y = {b Œ x2: b Œ W} Œ W, by SS, and look
at the statement x2 Õ y. If a counterexample exists then
there must a counterexample in W. This is a contradiction.
LEMMA 1.4. The following is provable in K(W). (x1 Œ W & x2 Õ
x1) Æ x2 Œ W.
We now use an axiomatization of ZF\Foundation by
extensionality, pairing, union, separation, power set,
reflection, and infinity. In the presence of separation,
pairing, union, power set take the form of the existence of a
set containing all of the requisite elements, rather than
being exact. Reflection takes the form:
($ transitive x1)(x2,…,xn Œ x1 Ÿ ("xn+1,…,xm Œ x1)(($xm+1)(j) Æ
($m+1 Œ x1)(j))), where m > n ≥ 2 and j is a formula in L(Œ)
whose free variables are among x2,…,xn.
LEMMA 1.5. Pairing is provable in K(W).
LEMMA 1.6. Union is provable in K(W).
LEMMA 1.7. Separation is provable in K(W).
LEMMA 1.8. Power set is provable in K(W).
LEMMA 1.9. Reflection is provable in K(W).
LEMMA 1.10. Infinity is provable in K(W).
LEMMA 1.11. K(W) proves ZF(rfn)\Foundation.
We now want to discuss interpretations of K(W).
The most natural models of K(W) are of the form
(V(q),Œ,V(k)), where q is a strongly inaccessible cardinal
and k < q is chosen with some care. Here it suffices to
simply choose k < q so that V(k) is an elementary submodel of
V(q). 5 By adapting this construction:
LEMMA 1.12. Every sentence in L(Œ) that is provable in K(W)
is provable in ZF(rfn)\F.
Here V(q) is replaced by V, and W is constructed via
Reflection. Thus we have:
THEOREM. A sentence in L(Œ) is provable in K(W) iff it is
provable in ZF(rfn)\F.
We now move on to K1(W), which is connected to indescribable
and subtle cardinals.
Let K1(W) be the following theory in the language L(Œ,W).
1. EXT.
2. SS.
3. RED+. (x1,…,xn Œ W & xn+1 Õ W & j) Æ ($xn+2,xn+3 Œ W)(xn+3 Õ
xn+1 Ÿ j[xn+1/xn+3]), where n ≥ 0 and j is a formula in L(Œ)
whose free variables are among x1,…,xn+2.
First of all, it is easy to see that RED+ implies RED by
taking xn+1 = W. Hence K1(W) extends K(W), and so we have
ZF(rfn)\F at our disposal.
We now review the relevant large cardinals in the treatment
of K1(W).
In ZFC, let n ≥ 1 and k be a cardinal. We say that k is nth
order indescribable if and only if for all R Õ V(k) and first
order sentence j, if (V(k+n),Œ,R) satisfies j, then there is
an a < k such that (V(a+n),Œ,R « V(a)) satisfies j.
We say that k is totally indescribable if and only if k is nth order indescribable for all n ≥ 2. This definition agrees
with the one given in [Ka94], p. 59.
It is natural to go a bit further.
In ZFC, we say that k is extremely indescribable if and only
if for all R Õ V(k) and first order sentence j, if
(V(k+k),Œ,R) satisfies j, then there is an a < k such that
(V(a+a),Œ,R « V(a)) satisfies j. 6
The definition of
to the definition
natural to simply
cardinal. Thus we extreme indescribability in ZF is identical
given above in ZFC except that it is
drop the requirement that the ordinal be a
speak of extremely indescribable ordinals. In ZF(rep)\F, we say that V(a) is strongly inaccessible iff
if the range of every function from an element of V(a) into
V(a) is an element of V(a).
LEMMA. ZF(rep)\F proves every strongly inaccessible rank
satisfies ZF; and a extremely indescribable Æ V(a) is
strongly inaccessible. $ an Œinterpretation of ZFC + “$ an
extremely indescribable cardinal” in ZF(rep)\F + “$ an
extremely indescribable ordinal” using constructibility.
THEOREM. K1(W) proves: ZF(rfn)\F + “$ an extremely
indescribable ordinal.” $ a standard model of ZF + “$ an
extremely indescribable ordinal.” $ an Œ model of ZFC + “$ an
extremely indescribable cardinal.”
Extremely indescribable cardinals fit into the standard large
cardinal picture as quite a bit stronger than weakly compact
cardinals, but quite a bit weaker than subtle or ineffable
cardinals, which are still compatible with the axiom of
constructibility. We now want to see what we need to build
standard models of K1(W).
In ZFC, let k be a cardinal. We say that f:k Æ S(k) is
regressive if and only if for all a < l, f(a) Õ a. (Here S is
the power set operation).
In ZFC, we say that k is a subtle cardinal if and only if
i) k is a cardinal;
ii) For all closed unbounded C Õ l and regressive f:k Æ
S(k), there exists a < b, a,b Œ C « k, such that f(a) = f(b)
« b.
Subtle cardinals are strongly inaccessible, and in fact
extremely indescribable.
THEOREM. ZFC + “there exists a subtle cardinal” proves the
existence of a standard model of K1(W); i.e., of the form
(V(k),Œ,V(l)). 7
Proof: Let k be a subtle cardinal. We want to find l < k such
that (V(k),Œ,V(l)) is a model of K1(W). You can’t just pick
any old elementary submodel V(l) of V(k) because Reducibility
has been strengthened considerably to RED+. The idea is to
define a map h from k into V(k) such that for each l < k,
h(l) is data sufficient for a counterexample to (V(k),Œ,V(l))
satisfying K1(W); default otherwise. This setup can be seen to
fit into the framework of the definition of subtle cardinal,
so that one obtains an appropriate a < b < k such that h(a)
and h(b) cohere. But then one can show that in light of this
coherence, the defaults must have been used, and so
(V(k),Œ,V(b)) is seen to satisfy K1(W).
Let K2(W) be the following theory in L(Œ,W).
1. EXT.
2. SS. x1 Œ W Æ ($x2 Œ W)("x3 Œ W)(x3 Œ x2 ´ (x3 Œ x1 &
j)), where j is a formula in L(Œ,W) in which x2 is not free.
3. RED. (x1,…,xn Œ W Ÿ j) Æ ($xn+1 Œ W)(j), where n ≥ 0 and j
is a formula in L(Œ) whose free variables are among x1,…,xn.
4. TRANS. ($x1 Œ W)(every transitive subset of W is a subset
of x1).
Note that SS is weaker than SS in that the quantifier x2 is
restricted to W. This is the natural formulation of Subworld
Separation that accommodates the possibility that W is not
transitive.
I.e., some elements of yesterday’s world may pick up new
elements from today’s world.
This form of Subworld Separation asserts that every property
of elements of any element of W has a representation, x2, in W
that at least represents the property with regard to elements
of W. We actually prove that this representation x2 Œ W is
unique.
Note that we can formally derive the nontransitivity of W.
For suppose W is transitive. Let W Õ x Œ W. By SS, let y Œ
W be such that for all z Œ W, z Œ x ´ z œ z. Then x Œ x ´
x œ x.
We have proved that K2(W) proves the existence of a standard
model of ZF + "there is a measurable rank." And ZF + "there
is a nontrivial elementary embedding from some rank into some
rank" proves the existence of a standard model of K2(W). 8 Measurable ranks are an appropriate formalization of
measurable cardinals in the ZF context with Choice. In ZFC,
these are equivalent: i.e., V(a) is a measurable rank if and
only if a is a measurable cardinal.
We say that V(a) is a measurable rank if and only if there is
a complete measure m on V(a+1); i.e.,
i) m:V(a+1) Æ {0,1};
ii) for all x Œ V(a+1), if x £ 1 then m(x) = 0;
iii) for all x Œ V(a+1), m(V(a)\x) = 1m(x);
iv) if x Œ V(a), f:x Æ V(a+1), and for all b Œ x,
m(f(b)) = 0, then m(»rng(f)) = 0.
We use x = y for ("z)(z Œ x ´ z Œ y).
LEMMA 4.1. Let j be a formula in L(Œ,W) which does not
mention x2. The following is provable in K2(W). x1 = x1. x1 =
x2 Æ x2 = x1. (x1 = x2 & x2 = x3) Æ x1 = x3. x1 = x2 Æ (j ´
j[x1/x2]).
For any formula j in L(Œ), let jW be the result of
relativizing all quantifiers to W.
LEMMA 4.2. Let j be a formula in L(Œ) whose free variables
are among x1,…,xn, n ≥ 0. The following is provable in K2(W).
x1,…,xn Œ W Æ (j ´ jW).
LEMMA 4.3. K2(W) proves Separation and Tupling.
Here Separation is as usual in set theory, except that we
accommodate all formulas in our language L(Œ,W). Tupling is
the scheme that allows us to form any {x1,…,xn}.
To get Separation, it suffices to show that Separation holds
in (W,Œ), because we can absorb W as a paremeter. But this is
clear by SS. So why didn’t we use Separation instead of SS?
This is because SS gives us something more: Separation in W
with respect to arbitrary formulas in L(Œ,W).
LEMMA 4.4. Let j be a formula in L(Œ,W) The following is
provable in K2(W). If x,y Œ W and every common element of x
and W lies in y, then x Õ y. x1 Œ W Æ [($!x2 Œ W)("x3 Œ
W)(x3 Œ x2 ´ (x3 Œ x1 & j)) & ($x2 Œ W « x1)("x3 Œ W)(x3 Œ x2
´ (x3 Œ x1 & j))]. 9 LEMMA 4.5. The following is provable in K2(W). No set includes
all ordinal as elements. The least ordinal, OW, that is not
in W is a limit ordinal > w. There is an ordinal in W\OW.
Fix l to be the first ordinal in W that is greater than OW. l
exists by Lemma 4.5.
LEMMA 4.6. The following is provable in K2(W). l is a limit
ordinal. If V(a) exists then V(a) Œ W ´ a Œ W. For all a <
OW, V(a) ŒÕ W. V(OW) exists and is a transitive subset of W.
V(l) exists.
Let W’ = {x Œ W: x Õ V(l)}.
LEMMA 4.7. The following is provable in K2(W). If x,y Œ W’
and x « V(OW) Õ y, then x Õ y. If x,y Œ W’ and x « V(OW) = y
« V(OW) then x = y.
LEMMA 4.8. The following is provable in K2(W). The
correspondence that sends x Œ W’ to x « V(OW) is oneone onto
V(OW+1). Every element of W’\V(OW) has rank l+1. The
correspondence is the identity on V(OW) and oneone from
W’\V(OW) onto V(OW+1)\V(OW). This correspondence is an Œisomorphism.
We are not claiming that this correspondence exists as an
actual function.
For x Œ V(OW+1), we let H(x) be the y Œ W’ such that x = y «
V(OW). Again, we emphasize that H does not necessarily exist
as a function.
Now define m:V(OW+1) Æ {0,1} by m(x) = 1 if V(OW) Œ H(x); 0
otherwise.
LEMMA 4.10. The following is provable in K2(W). V(OW) is
strongly inaccessible.
LEMMA 4.11. K2(W) proves: For all a < OW there exists a < b <
OW and a complete measure on some V(b+1).
THEOREM 4.12. K2(W) proves that there is a standard model of
ZF + “there exists arbitrarily large measurable ranks.” 10
We now wish to construct a standard model of K2(W) using large
cardinals.
In ZFC, we say that k is an extendible cardinal if and only
if for all a > k there is an elementary embedding from V(a)
into some V(b) with critical point k.
THEOREM 4.13. ZFC + “there is an extendible cardinal” proves
the existence of a standard model of K2(W).
Here a standard model is of the form (V(k),Œ,A), where A Œ
V(k) is the interpretation of W; A cannot be a rank.
Let k be an extendible cardinal. Let j:V(k+1) Æ V(a+1) be an
elementary embedding with critical point k. Then k and a are
strongly inaccessible and that V(k) is an elementary submodel
of V(a). Now let j’:V(a+1) Æ V(b+1) be an elementary embedding with critical point k. Then j’(a) = b, and since a is
strongly inaccessible, b is also strongly inaccessible. We
use j’ below for j’V(a).
Thus j’ is an elementary embedding from V(a) into V(b) with
critical point k. We prove that (V(b),Œ,rng(j’)) is a model
of K2(W). Clearly this is a legitimate interpretation since
rng(j’) Œ V(b), by the strong inaccessibility of b.
rng(j’) is an elementary submodel of V(b) by general nonsense
concerning elementary embeddings.
From this it follows that RED holds in (V(b),Œ,rng(j’)).
Also the largest transitive subset of rng(j’) is V(k). To see
this, suppose A is a transitive subset of rng(j’). Then there
are elements of A of every rank < rk(A). If rk(A) > k+1 then
some element of A, and hence some element of rng(j’), has
rank k+1. Therefore by RED, k Œ rng(j’), which contradicts
that k is the critical point of j’.
It remains only to verify SS.
Let j’(x) be given, where x Œ V(a). Let y Õ j’(x). Consider
j’(j’1[y]). Then j’(u) Œ j’(j’1[y]) if u Œ j’1[y] iff j’(u)
Œ y. Thus we have shown that for all v Œ W, v Œ
j’(j’1[y]) ´ v Œ y.
We can actually use quite a bit less. 11 THEOREM. ZFC + “there exists a nontrivial elementary
embedding from a rank into a rank” proves the existence of a
standard model of K2(W).
The use of an extendible cardinal actually proves the
standard model of an extension K3(W) of K2(W) by the
additional axiom of Reducibility:
RED’’. j Æ ($x1 Œ* W)(j), where j is in L(Œ) and x1 is not
free in j.
In K3(W), one can prove ZF(col)\F + “there exists a nontrivial
elementary embedding of a rank into itself”, which Woodin has
shown to imply Projective Determinacy. ...
View
Full
Document
 Fall '08
 JOSHUA
 Math, Set Theory

Click to edit the document details