This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
P, P' does not force Fa = Fp'. Again the symbol = is treated as before. (v) If a > /8, we reduce the case Fa t Fp to cases (in) and (it;) treated above. We say P forces F a « F 0 if for some 0' < fi, P forces i*> e F s and P forces ^ a = Fp> (i.e., (x)a(a; e F a ( = ) x e i^*) which is a statement of type (R and hence precedes Fa e Fp). We say P forces ~]FatFfi if for all 0' < 0 and P ' 3 P, P ' does not force both Fp' e Fp and Fp, = ^ a . The most important part of Definition 6 is I, the other parts are merely obvious derivatives of it. Definition 7: If a is an unlimited statement with r quantifiers, we define "P forces a" by induction on r. If r = 0, then a is a limited statement. If a = (x) h(x), P forces a, if for all P ' 3 P, and a, P' does not force 1 b(Fa). If a = Jx b(x), P forces a if for some a, P forces b(Fa). In the proofs of Lemmas 2, 3, 4, and 5, we keep the same well-ordering on limited statements as in Definition 6, and proceed by induction. LEMMA 2. P does not force a and ~~\ a, for any a and P. Proof: Let a be a limited statement with r quantifiers. If r > 0, and P forces both JaX b(x) and (x)a ~] b(x), then P must force b ^ ) for /? < a which means P cannot force (x)a ~\ b(x). Case II of Definition 6 will clearly follow from case III. Parts (i) and (ii) are trivial. If a is in part (Hi), then P forces a if and only if P
5 VOL. 50, 1963
MATHEMATICS: P. J. COHEN
1147
forces a statement of lower rank and in this case the lemma follows by induction. In part (iv), if P forces Fa which means P can not force ~\ Fa e Fp. In part (v) if P forces Fa t Fp, for some 0' < 0, P forces Fp> e Fp and Fa = Fp> which again violates P forcing ~i Fa e Fp. If a is an unlimited statement, the lemma follows in the same manner by induction on the number of quantifiers. LEMMA 3. If P forces a and P' ZD P, then P' forces a. Proof by induction as in Lemma 2. LEMMA 4. For any statement a and condition P, there is P' 3 P such that either P' forces a or P' forces ~~\ a. Proof: Let a be a limited statement with r quantifiers. If r > 0 and P does not force a = (x)a b(x), then for some P' Z) P, P' forces ~\ b(Fp), 0 < a, which means P' forces ~\a. If r — 0, we may restrict ourselves to III, for if we enumerate the components of a, by defining a finite sequence Pn, PB = P and P , + i D P , we may successively force each component or its negation so that finally either a or 1 o is forced. Again, cases (i) and (ii) are trivially disposed of. Case (in) is handled by induction as before. If a = Fa e Fp is in case (ID) then if P does not force ~\ a, for some P' => P and 0' < p, N(p') = 9, P' forces Fa = Fe. so P' forces a. If a == Fa e Fp is in case (v) if P does not force ~\ a, then for some P' Z) P, /3' < 0, P' forces Fp e Fp and Ffi- = Fa> hence P' forces a. Unlimited statements are handled as before. Definition 8: Enumerate all statements an, both limited and unlimited, and all ordinals an in 3E. Define Pzn as the first extension of Pin _ \ which forces either an or ~l an. Define P 2n + i as the first extension of Pin which has the property that it forces Fp t F«n where 0 is the least possible ordinal for which there exists such an extension of P2re, whereas if no such /3 exists, put P2« + i = PinThe sequence Pn is not definable in 3TC. Since all statements of the form n e as are enumerated, Pn clearly approach in an obvious sense, sets as of integers. With this choice of ae, let 91 be defined as the set of all Fa defined by Definition 1. LEMMA 5. All statements in 91 which are forced by some Pn are true in 91 and conversely. Proof: Let a be a limited statement with r quantifiers. If r > 0, then if P,, forces (x)a b(x), if /3 < a, then some Pm must force b(Fs) since no Pm can force ~| b(Fp). By induction we have that h{F0) holds, so that (x)a b(x) holds in 31. If Pn forces Jax b(x), for some /3 < a, Pn forces b(Ffi) so by induction b(F?) holds and hence laX b(x) holds in 91. Case I I will clearly follow from case I I I and (i) and (ii) arctrivial. If a is Fa e Fp or ~1 Fa e FB in case (in) then if Pn forces a, Pn forces precisely the statement which because of the definition of Fp is equivalent to a. In case (iv) if Pn forces Fa e Fp, for some /3' < 0, .V(j3') = 9, l\ forces F^' = F«, which therefore holds by induction in %.. If Pn forces ~| Fa t Fp, then for each p' < Ii, N(P') = 9, Fa = F 3 / is not forced by any P m so some P ra must force Fa ^ /'V' which proves ~| Fa t F0 holds in 31. Similarly for case (v) and for unlimited statements. Since every statement or its negation is forced eventually, the converse is also true. Lemma 5 is the justification of the definition of forcing since wo can now throw back questions about ill to questions about forcing which can bo formulated in 0??.
6 1148
MATHEMATICS: P. J. COHEN
PROC.
N. A. S.
In the next paper, we shall prove that, ill i^ a model for Z-F in which part 3 of Theorem 1 holds. * The author is a fellow of the Alfred P. Sloan Foundation. 1 Cohen, P. J., "A minimal model forset theory," Bull. Anier. Math. Soc, 69, 537-540 (1903). 2 Fraenkel, A., and Y. Bar-Hillel, Foundations of Set Theory (1958). 3 Godel, K., The Consistency of the Continuum Hypothesis (Princeton University Press, 1940). 4 Shepherdson, J. C , "Inner models for set theory," J. Symb. Logic, 17, 225-237 (1957). 6 Sierpinski, W., "L'hypothese gcneralisee du continu et l'axiome du ehoix," Fund. Math., 34 1-5(1947)
7
VOL. 51, 1964
MATHEMATICS:
THE INDEPENDENCE
P. J. COHEN
OF THE CONTINUUM HYPOTHESIS,
105
II*
BY PAUL J. COHEN1 DEPARTMENT OP MATHEMATICS, STANFORD UNIVERSITY
Communicated by Kurt Godel, November 27, 1963
This paper is a continuation of reference 1, in which we began a proof of the fact that the Continuum Hypothesis cannot be derived from the other axioms of set theory, including the Axiom of Choice. We use the same notation as employed in reference 1. THEOREM 2.
91 is a model for Z-F set theory.
The proof will require several lemmas. The first two lemmas express the princi-
8 106
MATHEMATICS:
P. J. COHEN
PROC. N. A. S.
pie that forcing is a notion which is formalizable in the original model 9TC. LEMMA 6. There is an enumeration aa of all limited statements by means of tlie ordinal numbers of 3TC, such that the usual formal operations -performed on statements are expressible by means of definable functions in 3TC of the indices a, for example, forming negations, conjunctions, replacing variables by partimdar sets, etc. Furthermore, the ordering corresponds to the definition of forcing given by trans/mile induction in Definition 6. LEMMA 7. Let o.{x,y) be a fixed unlimited statement containing two unbound variables x and y. The relation $a(P,«,/3) which says that ]' forces (i(Fa,Fp) and /3 is the least such ordinal, is definable in 3TC. This follows from the fact that using Lemma 6 the relation "P forces aa" can be formalized in Z-F as a statement about P and a. A given unlimited statement can also be handled since, after a finite number of replacements of variables, it is reduced to a limited statement. Definition 9: For a(x,y) as above, put To(a) = sup[/?|3 P,oti < a, $ 0 (P,ai,/8)}. LEMMA 8. Let a(x,y) be a fixed unlimited statement, a an ordinal. For each a' < a either there is no Fp such that a(Fa',F$) or such an F0 exists with /8 < r a ( a ) . Proof: If ,8 is the least ordinal such that a(Fa',Ffi), then a(Fa>,Fp) must be forced by some Pn which clearly implies (3 < ra(a). LEMMA 9. Let &(x,y) be an unlimited statement of the form QiXxQ&i,
• • -, Qnxnb(x,y,xh
. . . , x n)
where b has no quantifiers and Qt are either existential or universal quantifiers. In 31, assume <X defines y as a single-valued function of x. Then for each a tlicrc exist ordinals 7o, . . . , 7 r e such that for x e Ta, there exist y t Fyo such (hat a(x,y) and for (x,y) in Ta X Tyo, the statement a(x,y) holds if and only ifa(x,y) holds where a is the statement formed by restricting the quantifiers Qt in h to range over Fyi. Proof: Lemma 8 implies the existence of 70 such that for x t Ta, there is a y e Fyo such that a(x,y). Define yk by induction as follows: let gk(x,y,xu . . ., xk-\,z) be the condition (i) if Qk is universal, ~Qk+ixk+i,
(ii)
• • ., Qnx,,b(x,ij,xi,
. . ., xk_hz,xk+i
. . . xn)
or
if Qk is existential, Qk+ixk+x,
. . ., Qnxnb(x,y,Xi,
. . ., xk_hz,xk^l
. . . xn).
L e m m a 8 implies t h a t for some yk, for all {x,y,x,, . . ., xk^) e Ta X Fyo X . . . X Fyk_u either no z exists such t h a t gk{x,y,xu . . ., xk_hz) or t h e r e is such a z e Fyt. This clearly implies t h e lemma.
LEMMA 10. The Axiom of Replacement holds in 31. Proof: If a(x,y) defines y as a single-valued function of x in 31, then for any a if D = {x\jz, z t Fa& a(z,x)} then by Lemma 9, D is denned by a condition in which all variables are restricted to lie in fixed sets Fyi, which by the definition of the sets Fa implies that D is a set in 31. The only other axiom to verify which is nontrivial, is the Axiom of the Power Set. The proof we give follows closely the method in reference 2 used to prove that V = L implies the Continuum Hypothesis.
9 VOL. 51, 1964
MATHEMATICS: P. J. COHEN
107
LEMMA 11. Let W be a set in 3TC, consisting of conditions P, such that if Pi and Pt belong to W, then P\ U Pi is not an admissible condition {i.e., contains a contradiction). Then iV is a countable set (in 3TC). Proof: Define sequences nk and Pj as follows. Put n,i = 1 and P\ the first P in IT. (We assume the P are well-ordered.) If nk and P} for j < nk are defined, put Rk equal to the sot of till conditions n e as or ~ n e as such that they or their negations are contained in .some Pitj < nk. Let Pt, nk < i < nk+i, be finitely many P in W, such that for all I' in W, 3j, nk < j < nk+i and P and P , have precisely the same intersection with l{,:. This is possible since Rk is a finite set. We claim that W consists only of the I'). For if P t W, then since P is a finite set of conditions, and Rk Q /i\+i, tliere exists a k such that P fi Rk = P D fit+i. Let nk < j < nk+i, such that P , (1 / C /("t+i, / ' U Pj is an admissible condition, which contradicts the hypothesis. Definition 10: Put C(P,a) = ft if P forces ^ « P« and for P ' 3 P, 7 < 0, P ' docs not force /(1T e P^. If no such /3 exists, put C(P,a) = 0. The function C is definable in 9TC, by virtue of the general principle contained in Lemma G. LEMMA. 12. For any a, there are only couniably many (in 9TC) /? such that for some I', C(P,a) = p. Proof: For each such /3, pick one P such that C(P,a) = /3. Then the set of all such P must be countable by Lemma 10. LEMMA VS. Let S be an infinite set of ordinals in 31!. There exists a set S' of onlinalx, S' 3 .S', S' = S such that S' is closed under J(i,a,/3,y), Ki(a), C(P,a), I (a), far all I' and a, ft, 7 e S. Also a t S' implies a + 1 e S'. The statement S' = S, above, means that with respect to 3TC, the sets S and S' arc of the same cardinality. LEMMA 14. Let S be a set of ordinals closed under the operations in Lemma IS, and such that if a < 3KT, a e S. Then there is a map g mapping S 1-1 onto an initial segment of ordinals which preserves J, Kt, I, N, and such that if N(a) = 0 (or 9), g(a) = ft is the first ordinal such that N(ft) = 0 (or 9) and ft is greater than g(a') for a' < a. Also, g is the identity for a < S^ r . Proof: S and g in the lemma refer to sets in the model 3K. We define g by transfinite induction. For a < 3^ T , let g be the identity. If g is defined for all ft in S less than a, UI(a) = a 3 (i.e., N(a) = 0), putg(a) = sap{g(ft)\ft < aaud/3 e S]. If / ( a ) = ft < a, then if N(a) = 9 (i.e., a = ft + 1), put g(a) = g(ft) + 1. If i = N(u), 1 < i < 8, put g(a) = J(i,g(Ki(a)), g(K*(a)), g(ft)). One can now show by induction that if a e S, N(a) = 0, g maps the set of all ft < a onto an initial segment. The lemma then easily follows. LEMMA 15. If we put G(Fa) = F0(a) for a in S, then G is an isomorphism with respect to e of Ax = {Fa\a e S] onto A2 = {Fg(a)\a t S]. Proof: This follows by induction on a, in the same way as in 12.6 of reference 2. Observe that in examining the operations JF4 and EF5 we need the fact that if Fae Ai and is not empty, then it has a member in Ai preceding it. This is true since S is closed under C(P,a), and C(P,a) for some P is the smallest ft for which Fpt Fa, if Fa^ . LEMMA 16. / / Fp C Fa, then for some y, Fp = Fy, where y < 3 + X r in 3R.
10 108
MATHEMATICS.
P. J. COHEN
PROC. N. A. S.
Proof: Let S contain all 8 < a, all 8 < 3 ^ r and /3, and be closed under the operations in Lemma 13. Let g be the corresponding isomorphism. Then clearly g(S) = 8 if 8 < a. Thus, by Lemma 15, if we put y = g(/3), since Fp CZ Fa! Fy = F0. Since g maps <S onto an initial segment, "y < S and so the lemma is proved. LEMMA 17. The Axiom of the Power Set holds in 91. Proof: Since every subset of Fa is contained in Fp, where /3 is the first ordinal such that iV(/3) = 0 and ^ > a + XT, it is clear that the power set of Fa occurs in 91. This completes the proof that 91 is a model, the other axioms being trivially verified. Since rank Ff< a, 31 contains no new ordinals. LEMMA 18. / / N(a) = N(0) = 9, and Fa > Fp in 9ft, then Fa > Fp in 91. Proof: The point of this lemma is that ordinals do not change their relative cardinality in the model 91. The added complications in the definition of forcing due to N(a) — 9 are compensated for in the proof of this lemma, in that as a runs through the ordinals with N(a) = 9, Fa runs through the ordinals of 311 in a manner independent of the sequence Pn. More exactly, the map a -*• Fa is an orderpreserving map of the ordinals a, N(a) = 9, onto all the ordinals of 311. Thus assume that some element in 91 defines a relation 6+.
Since forcing is the natural method for producing independence results, set theorists have concentrated on a more specific form of the problem: Given a transitive model M of ZP +
GCH, can a positive solution be ob-
tained by forcing over M with a set of conditions IP ? This approach suggests a number of related problems: Is there a
IP which collapses 8 + to
6? Is there a IP which makes an inaccessible cardinal singular? Until very recently, there was a widespread assumption among set theorists that such sets of conditions do exist and"merely awaited discovery. Then Silver challenged this assumption by proving it false. Specifically, Silver proved - in ZPC - that if the continuum hypothesis holds below a singular cardinal f$ of uncountable cofinality, then it holds at 6. Thus, in many important cases, not only the narrower forcing problem but the general problem itself has a negative solution.
Much of the effort to produce a positive forcing solution centered on the attempt to exploit the properties of special ground models - either L or models containing large cardinals. The latter approach met with some success: Prikry, fr. ins., showed that a measurable cardinal can be turned into an ID cofinal cardinal. Magidor, starting with an elephantine cardinal, produced a model in which 2
n
< oi for n < ui and 2
w
> is + ..
Jensen's efforts to produce a positive solution over L led to total
.
14 116
K. Devlin & R. Jensen
failure. Silver's work then led him to consider the problem from a new perspective. He discovered that the statement " 0 * does not exist" (henceforth abbreviated as -i 0 * ) implies a negative solution to all cases of the singular cardinals problem. But then there cannot be a positive forcing solution over L, since every generic extension of L by a set of conditions satisfies -i 0
Throughout this paper we assume ZFC. Our main theorem says, in effect, that if -i 0
, then the "essential structure" of cardinalities and con-
finalities in L is retained in Theorem 1.
Assume -i 0
V.
. Let X be an uncountable set of ordinals. Then
there is a constructible set Y s.t.
X 2 by co. in Corollary 2. The following corollary establishes a totally negative solution of the singular cardinals problem over L. Corollary 3. (a)
Assume -i 0
. Let B be a singular cardinal. Then
B is singular in L
(b) e + = B + L (c)
If A c g s.t. Hg = Lg[A], then iP(B) c L[A].
(d)
cf(B) s Y < B
(e)
Let 0 = 2 ^ . Then R
—*• B Y = 2 Y • B +
feif\/Y « is suitable iff either J T ^— (There is a largest cardinal)
or else there are arbitrarily large y < x s.t. cf(y) > a> and J T
[=•= ( Y is
regular). § 0 Theorem 1 reduces to the statement: Lemma 1.
Let T i u>2 be a suitable cardinal in L. Let X c x be cofinal
in t s.t, I < T . Then there is Y 3 X s.t.
Y 6 L
and Y < x.
We first show that Lemma 1 implies § 0 Theorem 1. Suppose not. Let X be an uncountable set of ordinals for which the conclusion of §0 Theorem 1 fails. Choose x = lub(X) minimal for such X. Then X < 7, since otherwise the conclusion of § 0 Theorem 1 would hold with Y = x. Hence x > (»„. Now suppose that the conclusion of Lemma 1 held. There would then be Z € L s.t. X c Z
and t < T . Let p = z
and let f : p -
the
zn
standard code and the Z n standard parameter of a. We
recall the following facts: (1)
p n s the largest p s.t. <J , A) is amenable for all A € En(Ja) n 9 ( J p ) .
(2)
R c Jpn
is S n (J a ) iff R is
Z^J^-1
, A^ ).
(5) A° » P ° = 0 . (4) Let n * 1 and let h be the canonical S^ Skolem function for <J n-1 , R11'1). Pa 01
h"(w x J
Then p n is the least p s.t. J n-1 = p™ n a x (p}) for some p 6 J n and p a is the <j - least such p.
(5) R c J n is I j W n - i , A^~ ) in the parameter p^ iff R is rud in a a <J n , A > (i.e. R is the intersection of J n with a class rudimena a tary in A o ) '
18 120
(6)
K. D e v l i n & R. Jensen
L e t IT : <J— , A) —>•_ < J \ n , A ) ( i 2 0 ) . Then t h e r e i s a P **.: P a 1 a o » ^ s,t. s.t.
(T = p— , A" = A—. M o r e o v e r , J a and i r ( p - )
TT : J - — > z
unique
t h e r e i s a u n i q u e if z> IT
= p (j
£
n).
n+i All of these facts are established in [Dev] and [PS]. The next result, though not explicit in our reference articles, does indead follow easily from the above facts. Def 1
Let a £ 6, 0 & n s u>: »a is a E
•'••'•
—
there is no
sn(Jg)
cardinal {!„ regular) in Jo iff ri
'" •' ™
'
n
• '™~
——•
p
function mapping a subset of some y < ua onto (cofinal-
ly into) ma. a is a cardinal (regular) in Jg iff there is no f 6 Jg mapping a y < wet onto (cofinally into) ua. If a is a cardinal in J o and a 6 Ja s.t. a c J . p
p
cr
then <J a , a) is amenable.
Clearly, being a cardinal (regular) in J g is the same as being a ZQ cardinal (regular) in Jg.
Lemma 1.
Let n i 1, a s (3.
(i) If iiia is a I . cardinal but not a Z cardinal in J o , then n n 1 n p o < a s p_ . Moreover up. is the least y < oa s.t. there is 1 (J o ) p S B n p map of a subset of y onto a>a. (ii) If p. < a s p P
and a is regular in J.n-1 , then cf(ua) = P
PQ
P
cf(o ) p n " 1 ). Proof. Q
(i)
Pg = B i a- Using C O , (2) and the fact that for any p there is a
A^(J ) map of up onto J , we get: p g a a for i < n (by induction on i ) . Hence Pg
2 • a v s o , since f
U f =f •
c f . Finally, sup a
= wo since
QED
V Carrying the proof of Lemma 1 (ii) a step further, we get the following rather technical lemma which will be of service to us in § 5.
Lemma 2.
n-l n Let p Q a o > p. where a>a is regular in J o . Let X = cf ( w ) .
Then there is a sequence ir, p i a, A c J— s.t. p g £ rng(ir) and
20
122
(b) (c) (d)
K. Devlin & R. Jensen _ n-1 _ n-1 There is a unique F s . t , p = p^- , A = A^ . n _ p^- < a . n n n n If T ( P ^ ) = Pg > then ~(p^-) = P g .
Proof. We first prove the existence part of (a). Let p, A, p, Y , ) •*
>- Vi < a) T-(4,,«i,x>,<x>>).
But the last equivalence is expressible as a H- statement in <J—, A") (since Vi < to T"(,5>) = P
serving and rng(3') n J
= rng(c).
hence by the fact that a is Z. pre-
Hence p— ^ < p .
Contradiction!
QED (Fact 6) Facts it, 5, 6 immediately give:
Fact 7.
A = An
R
All that remains to be proved is Fact 8.