[to appear in the Journal of Symbolic Logic]

Superdestructibility: A Dual to Lavers Indestructibility Joel David Hamkins Saharon Shelah

Abstract. After small forcing, any <-closed forcing will destroy the supercompactness and even the strong compactness of .

modified:1998-10-05

In a delightful argument, Laver [L78] proved that any supercompact cardinal can be made indestructible by <-directed closed forcing. This indestructibility, however, is evidently not itself indestructible, for it is always ruined by small forcing: in [H96] the rst author recently proved that small forcing makes any cardinal superdestructible; that is, any further <-closed forcing which adds a subset to will destroy the measurability, even the weak compactness, of . What is more, this property holds higher up: after small forcing, any further <-closed forcing which adds a subset to will destroy the -supercompactness of , provided is not too large (his proof needed that < + , where the small forcing is <-distributive). In this paper, we happily remove this limitation on , and show that after small forcing, the supercompactness of is destroyed by any <-closed forcing. Indeed, we will show that even the strong compactness of is destroyed. By doing so we answer the questions asked at the conclusion of [H96], and obtain the following attractive complement to Laver indestructibility: Main Theorem. After small forcing, any <-closed forcing will destroy the supercompactness and even the strong compactness of . We will provide two arguments. The rst, similar to but generalizing the Superdestruction Theorem of [H96], will show that supercompactness is destroyed; the second, by a dierent technique, will show fully that strong compactness is destroyed. Both arguments will rely fundamentally on the Key Lemma, below, which was proved in [H96]. Dene that a set or sequence is fresh over V when it is not in V but every initial segment of it is in V . The rst authors research has been supported in part by the College of Staten Island and a grant from The City University of New York PSC-CUNY Research Award Program. The second authors research has been supported by The Israel Science Foundation, administered by the Israel Academy of Sciences and Humanities. This is publication 618 in his independent numbering system.

618

revision:1998-10-05

2 Key Lemma. Assume that |P| = , that P Q is -closed, and that cof() > . Then P Q adds no fresh subsets of , and no fresh -sequences. While in [H96] it is proved only that no fresh sets are added, the following simple argument shows that no fresh sequences can be added: given a sequence in , code it in the natural way with a binary sequence of length , by using many blocks of length , each with one 1. The binary sequence corresponds to a subset of the ordinal , which, since cof() = cof(), cannot be fresh. Thus, the original -sequence cannot be fresh. Let us give now the rst argument. We will use the notion of a -club to extend the inductive proof of the Superdestruction Theorem [H96] to all values of . Theorem. After small forcing, any <-closed forcing which adds a subset to will destroy the -supercompactness of . Proof: Suppose that |P| < and P Q is <-closed. Suppose that g G P Q is V -generic, and that Q = Qg adds a new subset A , with minimal, so | that A V [g][G] but A V [g]. By the closure of Q, we know that cof() . Suppose, towards a contradiction, that is -supercompact in V [g][G]. Let P denote (P )V [g][G] , which is also (P )V [g] . Lemma. Every normal ne measure on P in V [g][G] concentrates on (P )V . Proof: Let us begin with some denitions. Fix a regular cardinal such that |P| < < . A set C P is unbounded i for every P there is C such that . A set D P is -directed i whenever B D and |B| < then there is some D such that for every B. The set C is -closed i every -directed D C with |D| < has D C. Finally, C is a -club i C is both -closed and unbounded. Claim. A normal ne measure on P contains every -club. Proof: Work in any model V . Suppose that C is a -club in P and that is a normal ne measure on P . Let j : V M be the ultrapower by . It is well known that j " is a seed for in the sense that X j " j(X) for X P . By elementarity j(C) is a -club in M and j " C j(C). (We know j " C M because M is closed under < sequences in V .) Also, it is easy to check that j " C is -directed. Thus, by the denition of -club, we know (j " C) j(C). But (j " C) =
C

revision:1998-10-05

modified:1998-10-05

j() =
C
Claim

(j " ) = j " .

Thus, j " j(C) and so C

618

3 Now let C = (P )V . We will show that C is a -club in V [g][G]. First, let us show that C is unbounded. If P in V [g][G], then actually V [g], and so = g for some P-name V . We may assume that [[ || < ]] = 1 and consequently { | [[ ]] = 0 } C; so is covered as desired. To show that C is -closed, suppose in V [g][G] that D C has size less than and is -directed. We have to show that D C. It suces to show that D V since C = P V . Since Q is <-closed, we know that D V [g], and thus D = Dg for some name D V . In V let Dp = { C | p D }. It follows that D = pg Dp . There must be some p g such that Dp is -conal in D; for if not, then for each p g we may choose p D such that Dp contains no supersets of p . Since D is -directed and |g| < there is some D such that p for all p g. But must be forced into D by some condition p g, so Dp for some p g, contradicting the choice of p . So we may x some p g such that Dp is -conal in D. But in this case Dp = D and since Dp V we conclude D V . Thus C is a -club in V [g][G], and the lemma is proved. Lemma Let us now continue with the theorem. Since is -supercompact in V [g][G] there must be an embedding j : V [g][G] M [g][j(G)] which is the ultrapower by a normal ne measure on P . Lemma. P ()M = P ()V .
modified:1998-10-05

Proof: (). By the previous lemma we know that (P )V and so j " j((P )V ) = (P )M . Since M is transitive, it follows that j " M . And obtaining this fact was the only reason for proving the previous lemma. Now if B and B V then j(B) M , and since B is constructible from j(B) and j " it follows that B M as well. (). Now we prove the converse. By induction we will show that P ()M V for all . Suppose that B and B M and every initial segment of B is in V . By the Key Lemma it follows that B V unless cof() < . So suppose cof() < . By the closure of Q we know in this case that B V [g] and so B = Bg for some name B V . We may view B as a function from to the set of antichains of P. Since B may be coded with a subset of , we know B M by the previous direction of this lemma. Thus, both B and B are in M and g is M -generic. Since B = Bg in M [g] there is in M a condition p g such that p B = B. That is, p decides every antichain of B in a way that makes it agree with B. Use p to decide B in V and conclude that B V . This completes the induction. Lemma

618

revision:1998-10-05

4 Now we are nearly done. Consider again the new set A such that A V [g][G] | but A V [g]. Since j is a -supercompact embedding, we know A M [g][j(G)]. Since the j(G) forcing is <j()-closed, we know A M [g]. Therefore A = Ag for some name A M . Viewing A as a function from to the set of antichains in P, we can code A with a subset of , and so by the last lemma we know A V . Thus, A = Ag V [g], contradicting the choice of A. Theorem Corollary. By rst adding in the usual way a generic subset to and then to , where cof() > , one destroys all supercompact cardinals between and . In fact, one does not even need to add them in the usual way. This is because the proof of the theorem does not really use the full <-closure of Q. Rather, if P has size , then we only need that Q is -closed and adds no new elements of P . Thus, we have actually proved the following theorem. Theorem. After any forcing of size < , any further -closed forcing which adds a subset to but no elements to P will destroy the -supercompactness of . This improvement is striking when is small, having the consequence that after adding a Cohen real, any countably-closed forcing which adds a subset to some minimal destroys all supercompact cardinals up to . Let us now give the second argument, which will improve the previous results with a dierent technique and establish fully that strong compactness is destroyed.
modified:1998-10-05

Theorem. After small forcing, any <-closed forcing which adds a -sequence will destroy the -strong compactness of . Proof: Dene that a cardinal is -measurable i there is a -complete (non + complete) uniform measure on . Necessarily cof(). This notion is studied in [K72]. Lemma. Assume that |P| < , that Q adds a new -sequence over V P , minimal, and that is -measurable in V PQ . Then P Q must add a fresh sequence over V .

revision:1998-10-05

In V [g], let T = { u ord< | [[ u s ]]Q = 0 }. Thus, under inclusion, T is a tree with many levels, and Q adds the -branch s. For u T , let bu = [[ u s ]]Q .

Proof: This lemma is the heart of the proof. Assume the hypotheses of the lemma. So PQ s is a -sequence of ordinals not in V P , and is a -complete uniform measure on . Without loss of generality, we may assume that P Q is a complete boolean algebra on an ordinal. Suppose now that g G is V -generic for P Q. Let Q = Qg , and s = sgG .

618

5 Thus, bu is an ordinal. Let I = { (u), bu | u T }, where (u) denotes the length of u, and dene , bu , bu when < and bu Q bu . Since u v (u), bu (v), bv it follows that T, I, , and consequently I = is also a tree, under the relation , with many levels. Furthermore, the th level of I consists of pairs of the form , . For p P let us dene that a p b when p a b. Thus, = pg p . In V [g][G] let b = , bs . Thus, b I, and if < then b b and so there is some r g such that b r b . Since there are fewer than many such r, for each there must be an r which works for -almost every . But then again, since there are relatively few r, it must be that there is some r g which has this property for -almost every . So, x r g such that for -almost every , for -almost every , we have b r b . Fix also a condition p0 , q0 g G forcing r to have this property. Let t = b | < & for -a.e. , b r b . Thus, t is a partial function from to pairs of ordinals, and dom(t) . In particular, dom(t) is unbounded in . We will argue that t is fresh over V . First, notice that t | knowing t we could read o the branch s. Thus, t V .
|

V [g] since in V [g]

618

revision:1998-10-05

Nevertheless, we will argue that every initial segment of t is in V . Suppose < , and let t = t . By the minimality of it follows that t V [g], and so there is a P-name t and a condition p1 , q1 g G, stronger than p0 , q0 , forcing | this name to work. Assume towards a contradiction that t V , and that this is forced by p1 . Then, for each r P below p1 we may choose r < such that r does not decide t(r ) (or whether r is in the domain of t). But, nevertheless, for each r either for -almost every , b r br or else for -almost every , b r br (but not both). In the rst case it follows that t(r ) = br , and in the second it follows that | r dom(t). Since there are relatively few r, by intersecting these sets of we can nd a single which acts, with respect to the r , exactly the way -almost every acts. Fix such a . Thus, for each r we have either b r br , and consequently | t(r ) = br , or else r dom(t) (but not both). Notice that and b are just some particular ordinals. Fix some condition p , q below p1 , q1 forcing and b to have the property we mention in the sentence before last. Now we will argue that this is a contradiction. Let = p . There are two cases. First, it might happen that b r , for some ordinal . Such a situation can be observed in V . In this case, p , q forces = bs and therefore, by the assumption on , it also forces t() = , . Since t is a P-name, it follows that p t () = , ,

modified:1998-10-05

6 contrary to the choice of = p . Alternatively, in the second case, it may happen that b r , for every . In this case, by the assumption on , it must be | that p , q forces that dom(t). Again, since t is a P-name, it follows that | p dom(t ), contrary again to the choice of = p . Thus, in either case we reach a contradiction, and so we have proven that P Q must add a fresh sequence. Lemma Lemma. If cof() and is -strongly compact, then is -measurable. Proof: Let j : V M be the ultrapower map witnessing that is -strongly compact. By our assumption on cof(), it follows that sup j " < j(). Let = (sup j " ) + , and let be the measure germinated by the seed . That is, X i j(X). Since < j() it follows that is a measure on . Since j() < for all < it follows that is uniform. Since cp(j) = it follows that is -complete. For < , let B = { | < cof() < }. Since cof() = in M , it follows that j(B ) and consequently B for every < . Since B = , it follows that is not + -complete, as desired. Lemma Remark. Ketonen [K72] has proved that if is -measurabile for every regular above , then is strongly compact. This cannot, however, be true level-by-level, since if < are both measurable, with measures and , then is a complete, non-+ -complete, uniform measure on . Thus, in this situation, will be -measurable, even when it may not be even + -strongly compact. But the previous lemma establishes that the direction we need does indeed hold level-bylevel. Let us now nish the proof of the theorem. Suppose that V [g][G] is a forcing extension by P Q, where |P| < and Q is <-closed. Let be least such that Q adds a new -sequence not in V [g]. Necessarily, and is regular. By the Key Lemma V [g][G] has no -sequences which are fresh over V . Thus, by the rst lemma is not -measurable in V [g][G]. Therefore, by the second lemma, is not -strongly compact in V [g][G]. Theorem So the proof actually establishes that after small forcing of size < , any -closed forcing which adds a new -sequence for some minimal , with , will destroy the -measurability of . This subtlety about adding a -sequence as opposed to a subset of has the following intriguing consequence, which is connected with the possibilities of changing the conalities of very large cardinals.

618

revision:1998-10-05

modified:1998-10-05

Bibliography

Corollary. Suppose that is -measurable. Then after forcing with P of size < , any -closed Q which adds a -sequence, but no shorter sequences, must necessarily add subsets to . Proof: Such forcing will destroy the -measurability of . Hence, it must add subsets to . Corollary

Bibliography
[H96] Joel David Hamkins, Small Forcing Makes Any Cardinal Superdestructible, Journal of Symbolic Logic (in press) [K72] Jussi Ketonen, Strong Compactness and Other Cardinal Sins, Annals of Mathematical Logic 5 p. 47-76 () [L78] Richard Laver, Making the Supercompactness of Indestructible Under -Directed Closed Forcing, Isreal Journal Math 29 p. 385-388 ()

618

revision:1998-10-05

modified:1998-10-05