AN INTRODUCTION TO APPLICATIONS OF ELEMENTARY SUBMODELS TO TOPOLOGY

ALAN DOW
Abstract. The purpose in writing this paper is to demonstrate that the no-

tion of an elementary submodel is a simple but powerful tool which should be brought into the mainstream of set-theoretic topology. The hope is that this can be accomplished simply by supplying the reader with proofs of results, ranging from very trivial ZFC results to some quite di cult independence results, in which elementarysubmodelsplay a key role. The topologicalquestions which are discussed concern, for the most part, metrizable and rst countable spaces. However one new result in the paper is a proof from PFA that compact spaces with countable tightness have some points of rst countability.

This paper is an expanded version of the author's talk given at the Spring Topology Conference in Gainseville. The main purpose of both the talk and the paper is to give examples to demonstrate the usefulness of elementary submodels to settheoretically oriented topologists. The author is not alone in believing that elementary submodels should become as familiar a part of the language of set-theoretic topology as is the pressing-down lemma for example. I believe that, for set-theoretic topologists, elementary submodels provide: 1. a convenient shorthand encompassing all standard closing-o arguments; 2. a powerful technical tool which can be avoided but often at great cost in both elegence and clarity; and 3. a powerful conceptual tool providing greater insight into the structure of the set-theoretic universe. I hope to convince some readers of the validity of these points simply by (over)using elementary submodels in proving some new and old familiar results. This paper is not a survey of their use nor an adequate (or even rigorous) introduction to the concept { it is intended solely as a demonstration of how useful they can be even in some rather unexpected applications. The author's primary reference is Kunen's text K] and the reader is directed there for both an introduction and to discover what I probably should really have said in many of the proofs and discusssions. There are two new results worth mentioning in the paper. The rst is that it follows from the consistency of large cardinals that it is consistent that nonmetrizability re ects in the class of locally-@1 spaces. This result is similar to Fleissner's results in F] about left-separated spaces with point-countable bases. The second is that it follows from PFA that each compact space of countable tightness necessarily contains points of countable character. The second result
1991 Mathematics Subject Classi cation. 54A35. Key words and phrases. elementary submodels, metrizable, countable tightness, re ection. 1

Introduction

ALAN DOW

is related to a question of Arhangel'skii A2] and is just something that Fremlin, Nyikos and Balogh \missed" in the papers Fr], FrN] and B]. In the rst section we will introduce elementary submodels and establish some of the non-standard assumptions we will make in the remainder of the paper. In the three sections following we apply elementary submodels in increasingly di cult arguments. Most of the results in these sections concern metric spaces and the remainder are concerned with spaces of countable tightness. None of the results in these sections involve forcing or large cardinals (although their existence is acknowledged). Section ve concerns applications of elementary submodels to forcing arguments. Not surprisingly this is an area in which elementary submodels are particularly useful { especially when proper forcing is involved. The last two sections discuss large cardinals and iterated forcing respectively. For a set or class M and a formula, ' , in the language of set theory, the formula 'M is de ned recursively (see IV of K]). 'M is just the formula you get when you \restrict all the quanti ers to M ". However note that (x y)M is really (8a 2 M )(a 2 x ) a 2 y) , since (x y) is not in the language of set theory. However it does not take long for one to become accustomed to the meaning of 'M especially when M is a \model" of most of ZF . We say that M is a model of ' (denoted M j= ') if 'M holds .
if
i Preliminaries

De nition: If fa1; : : : ; ang M N then '(a1 ; : : : ; an) is absolute for M; N

'M (a1 ; : : : ; an) holds
i

'N (a1; : : : ; an) holds.

De nition: M is an elementary submodel of N , denoted M N , if M N and for all n < ! and formulas ' with at most n free variables and all fa1; : : : ; ang M the formula '(a1 ; : : : ; an) is absolute for M; N .
For a cardinal , the set H ( ) is the set of all \hereditarily < sized sets". That is, H ( ) is the set of all sets whose transitive closure has size less than . These sets are useful because if is regular then H ( ) j= ZF - P (see IV in K]). In practice, when one is investigating a property of some objects, say hX; T ; Ci, one usually knows the largest possible size of any set at all relevant to the validity of the property. Therefore there is a cardinal large enough and a formula of set theory ' so that '(X; T ; C ) expresses the property and such that '(X; T ; C ) is absolute for V; H ( ) . (For example, see the Levy Re ection Theorem, IV in K]). Throughout this paper we shall often choose such \large enough" or H ( ) with little or no discussion as to how large it needs to be. Once we have shrunk our model to a set (namely H ( )), we then have the downward Lowenheim-Skolem theorem. The proof of this theorem makes very transparent the concept of elementary submodels.

ELEMENTARY SUBMODELS

Theorem 1.1: For any set H and X H , there is an elementary submodel M of H , such that X M and jM j jX j !.
M is called an elementary chain if it is a chain when ordered by . It is worth
noting that is a transitive order. Another very useful notion and resulting basic fact concerns elementary chains.

Theorem 1.2: If M is an elementary chain then M

M for all M 2 M.

H , then M is an elementary chain and M H .

Corollary 1.3: If M is a chain under inclusion of elementary submodels of S

Corollary 1.4: For uncountable regular cardinals and X 2 H ( ) with jX j < , f < : 9M H ( ) X M ; jM j < and M \ = g
is a closed and unbounded set (cub) in . Proof. Inductively build an S elementary chain of length , fM :

for a limit ordinal, M = fM : < g.

< g, so that

Note that, for regular cardinals , if M H ( ) has cardinality less than then M 2 H ( ). Therefore we could have built the elementary chain so that M 2 M +1 chain or { this will be called an elementary 2-chain. A continuous elementary S elementary 2-chain is one in which, for each limit , we have that M = < M . Another corollary to theorems 1.1 and 1.2 which we shall use frequently is the following. Theorem 1.5: For any regular 2! = c and any X H ( ) with jX j c, there is an M H ( ) so that X M ; jM j = c AND M ! M: A remark often made for its shock value is to suppose that M is a countable elementary submodel of H ( (2c )+ ) such that the reals R are in M . Then M j= (R is uncountable) and yet R \M is only countable. There's no paradox here: M thinks R is uncountable not R\M . Indeed the set R\M is not even in M so M can't think anything about it. The lesson here is that if M H , then 'M (m) () 'H (m) holds for elements of M , and that, in general, neither of the implications X 2 M ) X M , X M ) X 2 M hold. However in some case X 2 M does imply X M .

countable element of M is a subset of M .

Theorem 1.6: If M H ( ), regular, and 2 M is a cardinal such that M , then for all X 2 M with jX j , X is a subset of M . In particular, each

ALAN DOW

onto ! X . Since ; X are both in M , M j= 9f onto ! X . That is, (9f onto ! X )M holds, hence there is an f 2 M such that (f maps onto X )M . Now we are down to what is know as a 0sentence (see IV in K]) - these formulas are absolute in many circumstances; that is f \really" is a function from onto X . Indeed, M j= (f X ) so we show that f X as follows. M j= (f X ) really means M j= (:9x 2 f n X ) { hence H ( ) j= (:9x 2 f n X ). Similarly H ( ) j= (f is a function ) since M j= (8 2 )(8x; y 2 X ) (f( ; x); ( ; y)g f ) x = y). Also, of course, M j= (8x 2 X )(9 2 ) ( ; x) 2 f ). Finally, we show that x 2 M for each x 2 X follows from M . Indeed, let x 2 X . Since f is \really" onto, we may choose 2 such that ( ; x) 2 f . Now M j= (9y)( ; y) 2 f so choose y 2 M such that M j= ( ; y) 2 f . Clearly H ( ) \thinks" (realizes?) that x = y.
Proof. If jX j

, then H ( ) j= 9f

When one says \let hX; T i be a topological space" it is usually meant that T is the topology on X . However we shall mean that T is a base for a topology on X . As we shall see below this is much more convenient. Suppose hX; T i is a topological space in some H ( ). Our general procedure is to take some kind of submodel, M H ( ) (frequently an elementary submodel), such that hX; T i 2 M . We then consider the (generally much smaller) subset XM = X \ M . At this point there are two natural topologies to consider on XM . On the one hand we have the subspace topology generated by fU \ XM : U 2 T g. And on the other hand, if M j= T is a base for a topology on X plus some basic axioms, then we'd get the base TM = fU \ XM : U 2 T \ M g. In general, these give very di erent topologies on XM . For example, if X = ! and M is countable then, of course, hXM ; TM i is a countable metric space. However, it is by comparing these two topologies that we prove our re ection results. The game we play is to jump back and forth between M and H ( ), comparing what M \thinks" with what H ( ) \thinks". Most of the results in this article are what are known as re ection results. A re ection question in topology usually has the form \if a space X has property P , then what is the size of the smallest subspace Y which also has property P ?". However it is usually the case that P is the negation of a nice property. So one might rephrase the question as \if is a cardinal and X is a space such that every subspace of X of cardinality at most has P , then does this guarantee that X has P ?". We will adopt the following notation : If P is a class of spaces or a property (which de nes the class of spaces having that property) then for a space X (X; P ) = minfjY j : Y X and Y does not have property P g ; (where we assume the minimum of the empty set is 1). There are not too many re ection results that hold for the class of all topological spaces but, for example, if we consider the separation property T1 then for any space X we have (X; T1 ) 2 f2; 1g. Another less trivial example is that (X; rst countable) !1 for all X such that (X ) = !1 (but not for all X such

ELEMENTARY SUBMODELS

that (X ) > !) . We shall use such self-explanatory abbreviations for classes of spaces as ` = !', `w = !' and `t ' for ` rst countable', `countable weight' and `tightness at most ' respectively. The reader is, of course, referred to the Handbook of Set-theoretic Topology for all topological de nitions and basic facts. In this section we prove a few simple theorems as an introduction to elementary submodel arguments. ]<! . Of course we want to show that there are n < ! ; F 2 ]<! and I 2 ] so that jF j = n and F \ F = F for all ; 2 I with 6= . Let M be an elementary submodel of H ( + ) such that fF j < g is in M and jM j < . Let = sup(M \ ) and choose any 2 . We have found our n and F ; let n = jF j and F = F \ M . Then one notes that M j= 8 < 9 (jF j = n ^ (F \ (max(F f g) + 1) = F ): To see this, note that the set S = f j (9 2 )F \ = F g is an element of M . Furthermore 2 S hence M j= S is co nal in . It follows that we may pick, by induction on < , F so that F \ maxF = F for all < . Alternatively, we may choose an elementary chain fM : < g of elementary submodels of cardinality less than so that M = M0 and choose F 2 M +1 so that F \ M = F : In the next example we prove Arhangel'skii's famous result that the cardinality of a Lindelof rst countable space is at most c .
Example 2.1 The Delta System Lemma. Let be a regular cardinal and let fF j
ii Some Elementary Applications.

< g

Example 2.2: A Lindelof space with countable pseudocharacter and countable tightness has cardinality at most c .

Let T be a base for a Lindelof topology on X which has countable tightness and pseudocharacter. Let hX; T i 2 M H ( ) such that M ! M and is \large enough".
Claim. M \ X = X

Indeed, suppose not, and choose z 2 X n M .

Subclaim 1.
For each

y 2 X \ M 9Uy 2 T \ M

such that

y 2 Uy and z 2 = Uy .

Proof of Subclaim 1. H ( ) j= (9fUn : n < !g T such that fyg = \fUn : n < !g).

Therefore M is a model of this , so let fUn : n < !g 2 M be such that M j= fyg = \nUn . Now since fUn gn 2 M it follows that V j= fyg = \nUn , hence we may choose Uy as required.
Subclaim 2. X \ M
is closed (hence Lindelof)

ALAN DOW

able set Y X \ M so that x 2 Y . Fix a set fUn gn2! T exhibiting that X has countable pseudocharacter at x . Next choose, for each n 2 ! a collection fUn;m gm2! TSsuch that x 2 = fUn;m gm2! andX n Un m2! fUn;m g. It follows that Y n fxg = fYn;m : n; m 2 !g where, for each n; m 2 ! Yn;m = Y \ Un;m . But since fYn;m : n; m 2 !g is a countable collection of countable subsets of M , the collection and each member of it is an element of M . Now if x were not in M we would have M j= Y = fYn;m : n; m 2 !g whereas H ( ) j= x 2 Y n Yn;m for each n; m 2 ! : Now by subclaim 1 U = fUy : y 2 X \ M g forms an open cover of X \ M but not of X . By subclaim 2 , U has a countable subcollection W which still covers X \ M . Now W is a countable subset of M and therefore is an element of M . But this is a contradiction since M j= W covers X . The result now follows from the fact that we may assume that jM j = c.

Proof of Subclaim 2. Assume x 2 X \ M . By countable tightness, choose a count-

Proposition 2.3: If a space X with base T has a point-countable base and hX; T i 2 M H ( ) then T \ M is a base for each point of X \ M .
H ( ) with hX; T i 2 M . Since H ( ) j= hX; T i has a point-countable base and M is an elementary submodel, there must be a set B 2 M such that M j= B is a point-countable base for hX; T i : It is straightforward to check that absoluteness guarantees that B is a base for hX; T i (in H ( )). Also H ( ) j= B is point-countable since this follows from M j= 8x 2 X fB 2 B j x 2 B g is countable. Now let x be any point of X \ M and suppose B 2 B is a neighbourhood of x. Choose U 2 T and W 2 B so that x 2 W U B . Now choose y 2 W \ M which we may do since x 2 X \ M . Since B is point-countable and fy; Bg 2 M it follows that fS 2 B : y 2 S g M ; hence , in particular, fB; W g M . Furthermore, since U 2 T and W U B , it follows that M j= 9T 2 T such that W T B : Therefore there is a T 2 T \ M such that x 2 T B which was to be shown.
Proof. Let M

As we shall see later, the hidden strength of the previous result is that the base T is not assumed to be point-countable (recall our assumption that T denotes a base, not the whole topology). The next result uses some compactness in the topological sense to nd when T \ M is not a base at all points of X \ M . Proposition 2.4: Let hX; T i be a countably compact space which is an element of a countable elementary submodel, M , of some su ciently large H ( ). if T \ M is not a base for hX; T i then 9z 2 X \ M such that T \ M is not a base at z .

ELEMENTARY SUBMODELS

Proof. Clearly we may as well assume that X \ M is not dense in X , so choose any z 2 X n X \ M . Now if T \ M does contain a base for all points of X \ M then there is a cover U T \ M of X \ M whose union does not contain z . But now X \ M is countably compact and U is a countable cover of it (since T \ M is countable) . Therefore there is a nite subcover , say W U , of X \ M and hence of X \ M . But now W 2 M and M j= W = X while H ( ) j= z 2 = W.

The following non-trivial result is an immediate consequence of the previous two propositions.
countable base has a countable base.

Example 2.5 Miscenko's Lemma: A countably compact space with a a pointiii Elementary Chains and the !-covering property

As we saw in the proof of Arhangel'skii's theorem it is a very powerful assumption to have that your elementary submodel is \closed under ! -sequences". Also we cannot expect that countable elementary submodels can \trap" a great deal. Indeed a typical inductive construction usually carries through without much di culty through the countable limit ordinals (discounting the problems of \trapping" the uncountable sets). On the other hand most constructions have considerable difculty passing !1, so we can expect some non-trivial re ection by taking elementary submodels of cardinality !1 even in the absence of CH . A useful property, which can to some extent replace \closed under !-sequences", is the !-covering property. We shall say that a set M has the !-covering property if for each countable A M there is a countable B 2 M such that A B . If fM : 2 !1 g is an elementary 2-chain of countable elementary submodels of some H ( ) such that for each 2 !1 M 2 M +1 then clearly the union of the M 's is an !-covering elementary submodel of H ( ) of cardinality !1. In this section we shall present several proofs that use elementary submodels of cardinality !1 which satisfy the !-covering property. It can be shown that such elementary submodels are exactly those which are uncountable and are the union of an elementary 2-chain of countable elementary submodels.

Theorem 3.1: If every subspace of cardinality !1 of a countably compact space is metrizable, then the space itself is metrizable.
It is convenient to make a few preliminary remarks before actually proving the theorem. To give a slick proof using elementary submodels it seems to be necessary to rst prove that such a space is necessarily rst countable, or at least that we may assume that if there is a counterexample then there is a rst countable one. This can be done directly with relative ease | because of countable compactness a counterexample would have a subspace with density at most !1 which was also a counterexample. However it seems more appropriate to proceed by rst proving the following surprising result of Hajnal and Juhasz (the result for regular spaces was proven by Tkacenko Tk]). This was proven during their systematic study of cardinal functions on unions of chains of spaces which is very similar to investigating re ection properties of the cardinal functions. We state this result twice in order to recall our notation introduced in xI.

ALAN DOW

Proposition 3.2 J]: If every subspace of cardinality at most !1 has countable weight then the space itself has countable weight. Proposition 3.2 J]: For any space X , (X; w = !) > !1 implies w(X ) = !.
cardinality !1 . We must rst show that T \ M is a base for the subspace topology on X \ M . Indeed suppose x 2 X \ M and U is an open neighbourhood of x . Since (X; w = !) > !1 , X \ M n U has a countable dense subset D. Since M has the !-covering property we may choose a countable D0 2 M so that D D0 X . Now M j= w(D0 fxg) = ! hence there is T 2 T \ M such that x 2 T and T \ D0 U . So we now have X \ M n U = D D0 n T X n T , hence M \ T U as was to be shown. It now follows that there is a countable subset B of T \ M which is a base for X \ M since w(X \ M ) = ! and T \ M forms a base. We may suppose B 2 M by the !-covering property . But now M j= w(X ) = ! , hence the result follows by elementarity.
Proof of 3.1. Let T be a base for the topology on X and assume that hX; T i is not metrizable. Let hX; T i 2 M where M is an !1-sized, !-covering elementary submodel of some H ( ). We shall show that X \ M with the subspace topology is not metrizable; hence (X; metriz ) = !1. By 3.2, we know that X has a subspace Z with jZ j = !1 and w(Z ) > !. By elementarity, there is such a set Z in M , so assume Z 2 M . Since X is countably compact and w(Z ) is uncountable we know that Z is not metrizable | hence we may as well assume that X = Z . We may also assume that, for each x 2 X , Z fxg is metrizable, hence rst countable. Therefore M j= Z fxg is metrizable. If X is not regular at x then M will re ect this since Z 2 M is dense. Indeed, suppose U 2 T is a neighborhood of x such that V n U 6= ; for each neighborhood of x. By elementarity, we may assume that U 2 M . Assume though that x has a neighborhood such that V \ M U (i.e. X \ M is regular). Since Z fxg is rst-countable and in M we may choose W 2 M such that W \ Z V . Therefore W V U . Since W and U are both members of M , this is a contradiction since M j= V U while H ( ) j= V n U 6= ;. So we may assume that X is regular at x and therefore it follows that X is rst countable at x and T \ M contains a local base at x. Therefore it su ces to show that hX \ M; T \ M i is not metrizable. Let fM : < !1g be a continuous -chain of countable elementary submodels of M with hX; T i 2 M0 and whose union is all of M . For each 2 !1 , we have that 9x 2 X \ M such that T \ M does not contain a base at x. But since fX; T ; M g 2 M +1 there is in fact an x 2 M +1 \ X \ M such that T \ M does not contain a base at x. Finally, let us suppose that hX \ M; T \ M i has a point-countable base and obtain a contradiction to Proposition 2.3 . Let N be a countable elementary submodel of H ( ) such that each of X; M; T and fM : 2 !1 g are in N . Let = N \ !1 and consider a point x 2 M \ X \ M such that T \ M does not contain a neighbourhood base at x as discussed in the previous paragraph. But now N j= Proof. Let hX; T i 2 M where M is an !-covering elementary submodel of H ( ) of

ELEMENTARY SUBMODELS

M = fM : 2 !1 g hence (T \ M ) \ N = fT \ M : 2 g: Therefore (T \ M ) \ N does not contain a local base for x 2 X \ M \ N , which is the contradiction we seek. A noteworthy aspect of the above proof is the double usage of elementary submodels. That is we developed some of the properties of the model M and then put M itself into a countable submodel. Clearly one of the awkward things about the above proof is that we had to rst show that the space would have to be rst countable in order to deduce that T \ M yielded the subspace topology on X \ M . We shall now discuss the situation for re ecting countable character. It is easy to see that (X; = !) > !1 6) (X ) = !: Indeed remove the limit ordinals having co nality !1 from !2 + 1 and observe that this example shows that even (X; = !) > !1&X is countably compact 6) (X ) = !: Therefore we could not have proceeded directly in 3.1 . But for which spaces does (X; = !) > !1 imply rst countability?

Proposition J]: For compact spaces X , (X; = !) > !1 ) (X ) = !:

It makes sense to ask how much compactness you need to obtain the above result. A space is called initially !1 compact if every cover by !1 open sets has a nite subcover. This condition is, of course, equivalent to each of the conditions \there is no free closed lter base of size !1 " and \each set of size at most !1 has a complete accumulation point". Let us rst observe that this is how much compactness one needs to prove Arhangel'skii's result relating free sequences and countable tightness. Recall that a sequence fx : < g is called a free sequence of length if for each g. When we say < it is the case that fx : < g is disjoint from fx : free sequence we shall assume the length is !1 .

Proposition 3.3: If a countably compact space does not have countable tightness then it contains free sequences. In addition, for an initially !1 -compact space X , t(X ) = ! i X has no free sequences .
Note that 3.3 is actually a re ection type result as well since it has as an immediate Corollary the fact that (X; t = !) > !1 ) t(X ) = ! for all initially !1 -compact spaces.

10

ALAN DOW

Proposition 3.4: For initially !1 -compact regular spaces X , (X; = !) > !1 ) (X ) = !:

Proof. Let hX; T i be a regular initially !1-compact space such that (X; = !) > !1. By the remark following 3.3 we have that t(X ) = !. Let M be an !-covering elementary submodel of some H ( ) so that hX; T i 2 MandjM j = !1. It su ces to show that M j= (X ) = !. As in 3.2 it su ces to show that T \ M induces the subspace topology on X \ M . Let x 2 X \ M and Tx = fT 2 T j x 2 T g. Let U 2 Tx and suppose that T \ M n U 6= ; for all T 2 Tx \ M . Using initial !1-compactness we may choose \ z 2 fT \ M n U : T 2 Tx \ M g: Using t(X ) = !, choose a countable set D X \ M n U so that z 2 D . Again, by !-covering of M we can nd T 2 Tx \ M so that x 2 TandT \ D = ;. Now, since we are assuming that X is regular and T 2 M we may choose T 0 2 Tx \ M so that T 0 T , hence T 0 \ D = ;. This is a contradiction since z is supposed to be in T 0 \ D.

I do not know if one needs to assume that X is regular in the previous result. If there is a non-compact rst-countable initially !1 -compact space then there is an example to show that the assumption of regularity in 3.4 is necessary. On the other hand, it easy to see that one does not need to assume regularity if CH holds. Indeed, this is because under CH ( and it is consistent with :CH) that every initially !1 -compact Hausdor space of countable tightness is compact!

Proposition 3.5: Let hX; T i be an initially !1-compact Hausdor space of countable tightness. Then every maximal free lter of closed sets has a base of separable sets. Furthermore if CH holds then the space is compact.
Proof. Suppose that F is a maximal free lter of closed subsets of hX; T i. Let M be an !-covering elementary submodel of some appropriate H ( ) such that fX; T ; Fg 2 MandjM jT = !1 . If CH holds we assume in addition that M ! M . Choose any z 2 FM = fF \ M : F 2 F \ M g, which we may do since jF \ M j = !1. Let A 2 F \ M be arbitrary and, by countable tightness, choose a countable set D A \ M so that z 2 D. Since M has the !-covering property and A 2 M we may assume that D 2 M . Since z 2 F \ D = F \ D for each F 2 F \ M it follows that M j= D \ F 6= ; for each F 2 F : Therefore, by elementarity and the maximality of F , D 2 F , F has a base of separable sets. It also shows that fF \ M j F 2 F \ M g F , FM 2 F and furthermore that jFM j > !1 since F is a free lter and X is initially !1-compact. Now suppose that M is closed under !-sequences and that z 0 is any other point of FM . Let Uz andUz be disjoint neighbourhoods of z and z 0 . Let Dz = D \ Uz and Dz = D \ Uz . Now just as we showed that D was in F , the same proof shows that both Dz and Dz are in F since they are both in M . However this contradicts = Dz . that z 2 F \ M for all F 2 F \ M since z 2
0 0 0 0 0

ELEMENTARY SUBMODELS

11

One can prove even a stronger result than the above one but the proof does not bene t by the use of elementary submodels and can be proven by a simple induction of length !1 .

Proposition 3.5A Fremlin]: If hX; T i contains no free sequences then for each countably complete maximal lter F of closed sets and each set H 2 F + = fZ X : Z \ F 6= ; for each F 2 Fg there is a countable H 0 H so that H 0 2 F
However an interesting feature of the proof of 3.5 is that it gives us a pretty good idea of how the consistency results in both directions must go. For example to show that it is consistent with :CH we can imagine that M is an inner model of CH and there are more reals to be added. It must be the case that new subsets of X \ M are added which can serve as the pair Uz ; Uz mentioned above. There are a lot of properties that we can show the pair must have | for example they both meet every countable set in M whose closure is a member of F \ M and that M j= F is a countably complete lter. We then investigate which kinds of forcings which add reals could not possibly add such a pair. It turns out that Cohen forcing is such a forcing but we shall not give the details here. In sections 5{7 we shall prove the result, due to , Fremlin and Nyikos that assuming the Proper Forcing Axiom, each initially !1 -compact space of countable tightness is compact. As for the consistency of there being such spaces the above analysis indicates that we have to plan for those inner models of CH and be building a space in such a way that it is possible to add the necessary sets. This is still open.
0

Another question which suggests itself is whether or not we could replace `compact' in the character re ection result with `countably compact & countable tightness'. It turns out that if there are large cardinals then it is consistent that simply `countable tightness' will su ce and no compactness is necessary at all. This will be proven in section 6. However it is consistent that these two properties do not su ce.

Example 3.6: In the constructible universe, L, there is a countably compact space of countable tightness and uncountable character such that each subspace of cardinality !1 has countable character.
It is shown in DJW], that there is a family of functions ff : < !2g in L so that 1. f : ! ! for each < !2 2. < < !2 implies f < : f ( ) 6= f ( )g is nite 3. 8f : !2 ! ! 9 < !2 such that f < : f ( ) 6= f ( )g is in nite. For each < !2 , let A ;0 = f( ; m) 2 ! : m f ( )g and for n > 0 let A ;n = f( ; n + f ( )) : < g. By a straightforward `Ostaszewski-type' induction one can de ne a locally countable, locally compact topology on !2 ! so that for each < !2 with uncountable co nality the subspace ! is countably compact and furthermore ensure that for each n < ! the set A ;n is clopen. Next one de nes, just as in DJW] , a topology on X = fpg !2 ! by declaring that !2 ! is endowed with the above topology and U is a neighbourhood of fpg S providing p 2 U and 8 2 !29n 2 ! so that U fA ;m : n < m 2 !g.

12

ALAN DOW

iv More on metric spaces { Hamburger's question.

Peter Hamburger has asked a natural question about metric spaces which can be asked in our terminology as \Does there exist a rst countable non-metrizable space, X , such that (X; metriz ) > !1 ?". If the existence of large cardinals is inconsistent, then the answer is known to be \yes". In fact the example would just be a special kind of subspace of the ordinal space !2 { called an E-set . An E- set is what is known as a non-re ecting stationary set. A set E of ordinals is called an E-set if E is stationary in its supremum, (8 2 E ) cf ( ) < !1 and for each < sup(E ) with cf ( ) > ! E \ is not stationary in . As mentioned above if there are no large cardinals then in fact there is an E-set contained in !2 (see De2]). In section 6 we shall discuss the consistency, from a large cardinal, of there being no E-sets. Therefore Hamburger's question for ordinal spaces is resolved. We shall shall show that the situation is the same for locally-@1 spaces. Recall that X is locally- if every point has a neighbourhood of cardinality at most . We proceed by analyzing the inductive step: \if X is a locally small space, does (X; metriz ) ) (X; metriz ) > ?" The singular case holds in ZFC and the consistency of the regular case follows from (and implies) the consistency of large cardinals. The main tools will be Proposition 2.3 and elementary chains.

Theorem 4.1: Suppose ! = cf ( ) Then (X; metriz ) 6= .

< and that X is a locally- space.

Proof. We may as well assume that X has cardinality . Fix a base B for X

consisting of open sets of cardinality at most . Choose a regular cardinal much larger than and an elementary 2-chain fMn : n <S !g so that f ; hX; Big M0 H ( ), jMn j < for each n 2 !, and X Mn . By assumption, Xn = Mn \ is metrizable for each n 2 !. Furthermore, for each B 2 B \ Mn , Theorem 1.6 implies that B Xn - hence Xn is open in X . Therefore X has a point-countable base. Furthermore, B 2 B\ Mn ) B Xn , hence B \ Mn does not contain a base for any point of X n Xn . Also by 2.3, B \ M contains a base for all points of Xn . Therefore Xn is a clopen subset of X and fXn+1 n Xn : n 2 !g is a partition of X into clopen metrizable pieces.

Theorem 4.2: Suppose cf ( ) Then (X; metriz ) 6= .

Proof.

< and that X is a locally- space.

The former (published) proof of Theorem 4.2 is very wrong. I would like to thank Paul Szepticky for bringing this to my attention. In addition I am very grateful to Stavros Christodoulou for supplying a correct proof. The idea is to generalize the proof of Theorem 4.1. One must rst prove that Proposition 2.3 can be generalized by replacing \point-countable" by \pointlambda" and demanding that be a subset of M .

ELEMENTARY SUBMODELS

13

Proposition 2.3(a). If a space X with a base T has a point- base and hX; T i 2 H ( ) and M , then T \ M is a base for each point of X \ M .

fM : 2 cf( g be an as in the \proof" of 4.2. For each , let T be a pointcountable base for the open metrizable space M \ X . Then one can easily see that S T = T is a point-cf( )g base for X .
Now one nishes the proof by showing that each M \ X is closed just as in 4.1. `Locally- ' can be replaced by `locally-< ' in 4.1, but I don't know if it can be in 4.2. Also we leave as an open question, the regular cardinal version of 4.2.

Next one must show that an X as in 4.2 has a point- base. To see this, let

Question 4.3: If is regular and there is a locally-< topology on the set in which every subspace of cardinality less than is metrizable , does it follow that h ; T i is not metrizable i f < j cf ( ) = !and 6= g is stationary ?
The proof of the next result must be delayed until 6.1.

Theorem 4.4: If it is consistent that there is a supercompact cardinal then it is consistent that, for the class of locally-@1 spaces , (X; metriz ) > !1 ) X is metrizable
Recall that a space is said to be (@1 -)CWH (for Collection-Wise Hausdor ) if every (@1 -sized) discrete set can be separated by disjoint open sets. Shelah has also proven that it is consistent (subject to a large cardinal) that a locally-@1 rst countable space which is @1 -CWH is CWH. However when the local smallness condition in this and the above results on metric spaces are dropped no such re ection results are known to hold. It is known that the situation is di erent since an example in F] shows that 4.4 does not hold if local smallness is dropped. To nish this section we will rst formulate a combinatorial principal on !2 and then construct a space from it. I do not know whether or not this combinatorial principle is a consequence of GCH or even 2!1 = !2 | it is consistent with these assumptions. 0 be the co nality ! limits in !2 and let (y) denote the statement: Let S2 0 g and fg j 2 S 0 g so that: (y) 9fA j 2 S2 2 1. A is a co nal increasing sequence in ; 2. g is a function from A into !; 0 3. 8 < !2 9h : ! ! such that 8 2 \ S2 f 2 A j h ( ) g ( )g is nite; and 0 so that 4. 8g : !2 ! ! 9 2 S2 f 2 A j g( ) g ( )g is in nite.

Example 4.5: (y) implies there is a rst countable space which is @1 -CWH and for which subspaces of size @1 are metrizable but which is not CWH and not metrizable.

14

ALAN DOW

0 g and fg : 2 S2 0 g be as in (y) . We shall de ne a topology on Let fA : 2 S2 the set !2 !2 !2 ! so that !2 is closed discrete and unseparated and the rest of the space is open and discrete. 0 let f n : n 2 !g list A in increasing order. For each point For each 2 S2 2 !2 we de ne a countable neighbourhood base U ( ; n) as follows: 0 U ( ; n) = f g f g !2 (! n n) ; for 2 = S2 0 for 2 S2 U ( ; n) = f g f g !2 (! n n) f( m ; ; g ( m )) : n < m 2 !g. The simplicity of the space ensures that a subspace will be metrizable if and only if it is CWH. To see that the space is @1 -CWH , let < !2 and choose h as in (y) . 0 , de ne h0 ( ) = h ( ) + j where j is such that h ( n ) > g ( n) For each 2 \ S2 for all n > j . Otherwise de ne h0 equal to h . It is easy to check that this h0 as a function from into the neighbourhood bases yields a separation of . Let us now show that the space is not CWH. Indeed suppose that g : !2 ! ! is such that U ( ; g( )) is disjoint from U ( ; g( )) for each < < !2. Choose 0 so that A0 = fm 2 ! : g( ) g ( )g is in nite. Let m = g( ) and choose 2 S2 n n m < n 2 A0 . But now the point ( n ; ; g ( n) is in both the sets U ( n ; g( n)) and U ( ; g( )).

v Elementary Submodels in Forcing Proofs

Forcing, of course, is the technique developed by Cohen which takes a (ground) model of set theory, together with a `new' desired set, and canonically constructs a model of set theory (the extension) containing the new set and the ground model. The di cult part of most forcing arguments is to show what sets are not added. That is, one must prove some kind of preservation argument. For example, it is frequently important that the ordinal which is !1 in the ground model remains so in the extension | we would say that \!1 is preserved". Some other examples of properties which we may want preserved include: \being an ultra lter over !"; \a tree having no co nal branches"; \a Souslin tree remaining Souslin". If V is the ground model, and P 2 V is a poset then we assume the existence of G P { a generic lter (i.e. for each dense open D P with D 2 V G \ D 6= ;). V G] is just the model obtained by adding G to V and using the axiom of comprehension to interpret all the P -names from V . (The fact that this works
is remarkable and di cult to prove but to apply it is not as di cult as I suspect is commonly assumed ). Therefore we now have two models of set theory, V and

V G]. If T 2 V is a base for a topology on X 2 V , then we can still discuss hX; T i in V G] | it will be the same topological space but it may have di erent second order properties. That is, we would want to discuss the preservation of topological properties such as: the countable compactness of hX; T i, the non- normality of hX; T i , etc. . In this section we give some examples of how elementary submodels can be utilized in proving such preservation results. We begin with Cohen-real forcing. Recall that the poset Fn(I; 2) = fs : s is a function into 2 ,dom(s) 2 I ]<! g and is ordered by s < t () s t.

ELEMENTARY SUBMODELS

15

then both

Lemma 5.1: If G is Fn(I; 2)-generic over V and A 2 V G] is a subset of !,

FA = fB ! : B 2 V andA B g and IA = fB A : B 2 V g

are countably generated.

claim and that F is countably generated. That I is countably generated is proven analogously.

To prove the claim, let p0 = p \ M and B 0 = fn : 9q < p0 so that q j`n 2 _ _ B 0 . This proves the Ag. Clearly B 0 B and B 0 2 M . Furthermore p0 j`A

_ be a Fn(I; 2)-name for A and let A _ 2 M H ( ). Proof. Let A _ B , then 9B 0 2 M with p j`A _ B0 Claim: If B 2 V and p j`A

B:

Lemma 5.2: Suppose hX; T i is a space and x 2 X is such that t(x; X ) = ! then 1 j`Fn(I;2) t(x; X ) = !.
showing that
0

_ ) and A _2M Proof. Suppose 1 j`(x 2 A

H ( ). We shall complete the proof by

_ \ M: 1 j`x 2 A _ Assume that p j`U \ (A \ M ) = ; where x 2 U 2 T . Let p0 = p \ M and de ne _ g and note that x 2 Ap and Ap 2 M . Since Ap = fy 2 X : 9q < p0 q j`y 2 A M j= t(x; X ) = ! , x is in the closure of some countable subset B of Ap which is an element of M . Now choose y 2 U \ B and, by elementarity, p0 > q 2 M so _ . Finally, we have our desired contradiction since dom(q) \ dom(p) = that q j`y 2 A 0 _ dom(p ), hence p q 2 Fn(I; 2), and (p q) j`y 2 U \ A
0 0 0

Although countably closed forcing does not preserve countable tightness in general, it is often the case that additional hypotheses on the space are required to prove the desired preservation result.

Lemma 5.3: If X is a space of countable tightness which is rst countable on countable subsets then the countable tightness of X is preserved by countably closed forcing.
_ a P -name and assume p j`x 2 A _ for Proof. Let P be a countably closed forcing , A _ Xg 2 some x 2 X . Let M be a countable elementary submodel such that fp; x; A; M . With Aq de ned as above, we have that x 2 Aq \ M for each q 2 M \ P . By assumption X \ M is rst countable hence choose fUn : n 2 !g a neighbourhood base for x in the subspace X \ M . Within M choose a descending sequence fpn : n 2 !g P with p0 = p and for each n 2 ! there is an xn 2 X \ M so that _ . Finally since P is countably closed there is a q 2 P with pn j`xn 2 Un \ A _ . This completes the proof since q < pn for each n, and q j`fxn : n 2 !g A x 2 fxn : n 2 !g.

16

ALAN DOW

Another preservation result for Cohen forcing we'll need is 5.4. This result is proven in DTW] and we shall not give a proof here. The proof uses a combinatorial structure on the Cohen poset called an endowment and elementary submodels do not play a role. Proposition 5.4: If hX; T i is a space such that for some set I , 1 j`Fn(I;2) hX; T i has a -discrete base then X must already have one. A poset P is de ned to be proper S] if for each > ! the stationarity of each stationary S ]! is preserved by forcing with P . Recall that C ]! is closed if the union of each countable chain contained in C is again in C . The elementary submodel approach makes the concept of properness much easier to understand and to use. In fact properness can be viewed as a condition which guarantees that many elementary submodels in V will extend to elementary submodels in V G]. If is a large enough cardinal and if = jH ( )j we can identify ]! and H ( )]! . Furthermore the set of countable elementary submodels of H ( ) is closed and unbounded. Since P is proper , it can be shown that if G is P-generic over V , then in V G] the set fM 2 H ( ) : M H ( )andM \ V 2 V g is stationary in H ( )]! . Therefore there are \stationarily many" such M such that P; G 2 M . Now we have H ( ) j= G \ D 6= ; for each dense open D P such that D 2 V hence by elementarity M j= G \ D 6= ; for each dense open subset D of P such that D 2 V . It also follows that M \ V is an elementary submodel of the H ( ) of V . Any condition q 2 P which forces that G \ M meets each dense open subset from M \ V is called a (P; M \ V )-generic condition. Combinatorially, in V , this translates to q 2 P is (P; M )-generic if for each r < q and each dense open D 2 M there is a condition p 2 D \ M such that r is compatible with p. As a result there is an equivalent de nition of proper which is the one we shall work with. P is proper i for each regular > 2jP j and each countable elementary submodel, M , of H ( ) which includes P , there is a (P; M )-generic condition below each p 2 P \ M (see S]).

Lemma 5.5: If P is an !1-closed poset, then P is proper and furthermore, if X 2 V and G is P -generic over V then X ]! V .
Proof. Let fDn : n 2 !g list the dense open subsets of P which are in M - a countable elementary submodel. Let p0 = p be any element of P \ M and choose a descending sequence pn , n 2 ! so that pn 2 Dn \ M . Since P is countably closed, there is a q 2 P so that q < pn for each n 2 !. This q is clearly an (M; P )-generic condition. Furthermore, this q has the property that for each element of M which is a P -name of a function from ! into V , q forces it to equal a function in V This is how one proves X ]! V .

A useful generalization of countably closed forcing is the iteration of Cohen forcing followed by countably closed forcing. There are many preservation results for the iteration which do not hold for countably closed forcing itself.

ELEMENTARY SUBMODELS

17

_ is a Fn(I; 2)-name of a countably closed poset and Lemma 5.6: Suppose Q that I is uncountable. If hX; T i has countable tightness at x 2 X , then 1 j`Fn(I;2) QhX; T i has countable tightness at x :
Proof. By 5.2 we may begin by assuming that G is Fn(I; 2)-generic over V and

Proof of Claim. Let U 2 T be an open neighbourhood of x and let p 2 Q \ M . _ g. Since Ap 2 M and Recall the de nition of Ap = fy 2 X : 9p0 < p p0 j`y 2 A _ and B 2 M . M j= t(x; X ) = !, there is a countable B Ap such that x 2 A Therefore U \ B 6= ; and furthermore, by elementarity for each y 2 B there is a _ . This shows that p0 2 M so that p0 j`y 2 A _g DU = fp 2 Q \ M : 9y 2 U \ M such that p j`y 2 A

Now since I is uncountable and M is not, there are, in V G], lters on Fn(!; 2) which are generic over V G \ M ]. That is, if P is any countable atomless poset which is an element of V G \ M ], then there is a lter H P so that H 2 V G] and H \ D 6= ; for all dense open D P with D 2 V G \ M ]. Well, such a P is Q \ M and so we choose such an H Q \ M . Since Q is countably closed, choose q0 2 Q so that H fp 2 Q : q0 < pg. _g : Claim: x 2 fy 2 X \ M : q0 j`y 2 A

_ be a Q-name of a subset of X , q 2 Q , and assume that q j`x 2 hX; T i 2 V . Let A _ . Let M be a countable elementary submodel containing fX; T ; I; Q; q; A; xg. A

is a dense open subset of Q \ M . Furthermore U 2 V , hence U 2 V G \ M ] , and so H \ DU 6= ;. Since q0 is below every member of H , this completes the proof of the claim and the Lemma. The condition X ]! V in 5.5 gives us a kind of !- absoluteness for V relative to V G] which is similar to what we had when we were taking elementary submodels closed under !-sequences. For example we have the following result.

Lemma 5.7: If G is generic over a countably closed poset P and hX; T i 2 V is a countably compact space having no free !1 -sequences, then V G] j= hX; T i is countably compact, with no free !1-sequences .
example, there are no new countable subsets of X and each countable subset from V still has all its limit points. Now suppose that fx _ : < !1 g is a P-name so that 1 j`fx _ : < !1 g is a free sequence in X: Since P is countably closed we can choose, in V , a descending sequence fp : < !1g P and fy : < !1g so that , for each , p j`y = x _ . It follows that fy : < g \ fy : < g = ; , for each < !1 , since p j`y = x _ for each < . But now the sequence is fy : 2 !1g is a free sequence since, by 3.3, X has countable tightness in V .
Proof. It is a trivial consequence of 5.5 that X is still countably compact since, for

18

ALAN DOW

Todorcevic pioneered the use of elementary submodels as \side conditions" in building proper posets. The following result is due to Fremlin for some special cases and the general result is due to Balogh. Proposition 5.8: If hX; T i is a non-compact, countably compact space , then there is a proper poset Q so that j`Q hX; T i contains a copy of the ordinal space !1 :
Proof. It turns out that the proof splits into two essentially di erent cases, depend-

ing on whether or not X contains free sequences. As we only plan to use the case when X does not, we shall only prove the result for this case and refer the reader to B], or D] for a proof of the other case. Since the iteration of proper posets is again proper, we may assume, by 5.7, that we have already forced with the countably closed collapse of the cardinal jT j + jX j. Therefore we may assume that X has cardinality and character !1 . Choose a free maximal closed lter F and de ne Q as follows. q 2 Q if q = hgq ; Hq ; Mq i where: 1. Hq 2 T ]<! ; 2. Mq is a nite elementary 2-chain of countable elementary submodels of H ( ) such that fX; T ; Fg 2 M for each M 2 Mq ; 3. gq is a function whose domain, Eq , is a subset of f 2 !1 : 9M 2 Mq M \!1 = g; T 4. for each 2 Eq and each M; M 0 2 Mq , 2 M 0 n M ) gq ( ) 2 M 0 \ fF \ M : F 2 F \ Mg : The actual de nition of the conditions is designed to make the nding of (M; Q)generic conditions a triviality. Indeed, if > 2jP j and M H ( ) with P 2 M and p 2 M \ P , then, we will show below that, q = hgp ; Hp; Mp fM \ H ( )gi is (M; P )-generic. We take care to ensure that the range of the union of the rst coordinates over the generic lter will yield a copy of !1 by de ning q < p providing: gq gp ; Hq Hp ; Mq Mp and, \ gq ( ) 2 fU 2 Hp : gp ( ) 2 U g for each 2 Eq \ maxEp , where = min(Ep n ) : It is not too di cult to show that, if P does not collapse !1 and G is P -generic then !1 frange(gp ) : p 2 Gg hX; T i: The main di culty to this claim is in showing that 8p 2 G; 8M 2 Mp ; 9q 2 G; M \ !1 2 Eq : However anyone who reads the rest of the proof can easily do this. One may nd it easier to slightly change the de nition of the conditions by allowing Hq 2 T !1]<! and for q < p add the condition that > max(Hq \ ) for each 2 E result of this q \max(Ep ) and = min(Ep n ): The S S is that, if G is Q-generic, E = fEp j p 2 Gg is a cub in !1 and g = fgp j p 2 Gg is a homeomorphism. That is, if 2 Ep , then p j` 2 E and if p j` E \ has no maximum then p j` (E \ ) co nal in (keep adding things to Hp \ ) and

ELEMENTARY SUBMODELS

19

p j` fg( ) j 2 E \ g converges to g( ) - again keep adding neighbourhoods of g( ) to Hp \ T . However the hard part of the proof is to show that P is proper (hence preserves !1). Let > 2jP j and let P 2 M H ( ) with jM j = ! and let p 2 P \ M . De ne p0 = hgp ; Hp; Mq fM \ H ( )gi. We must rst show that p0 2 P and that p0 < p. First of all P 2 M and = supf : 9q 2 P; 9M 0 2 Mq 2 M 0g 2 M , hence M \ H ( ) H ( ). Furthermore, if M 0 2 Mp , then M 0 2 M and M j= M 0 H ( ) , hence M 0 M \ H ( ). It follows that p0 < p. Now consider r < p0 and D 2 M such that D is a dense open subset of P ; without loss of generality we may assume r 2 D. Let r0 = hgr \ M ; Hr \ M ; Mr \ M i ~ = fhgq ; Hq i : r0 q 2 Dg 2 and note that r0 2 P \ M and that r r0 . Let D ~ \ M such H ( ) \ M . Let us rst note that it su T ces to nd a pair hg; H i 2 D that range(g n gr ) U where U = fU 2 Hr : gr ( 0 ) 2 U g and 0 = min ~ \ M , then by elementarity there is q 2 D \ M so Er n M . Indeed, if hg; H i 2 D that hgq ; Hq i = hg; H i and r0 < q. One easily checks that q and r are compatible. Let Er n M = f 0 ; : : : ; n 1g listed in increasing order. For expository purposes, rst suppose that n = 1. Then, by de nition of r 2 P we know that \ gr ( 0 ) 2 fF : F 2 F \ M g : Now, by 3.5A , we may assume that F is just a base for the lter which consists of separable sets. Therefore F = F \ M for each F 2 F \ M and, since F is T ~ 2 M \ H ( ) , it follows countably complete fF : F 2 F \ M g 6= ;. Since D ~ that Z = fx 2 X : 9hg; H i 2 D fxg = range(g gr0 )g 2 M \ H ( ). Therefore if Z2 = F + , then there is some F 2 F \ M such that F \ Z = ;. But this contradicts that gr ( 0 ) 2 F \ Z . Therefore it follows that gr ( 0 ) 2 Z Hence we may choose ~ \ M such that fxg = range(gq such an x in U and a q ~2 D ~ gr0 ). The idea of the elementary chains is that we can then handle the case n > 1. For i = 0; 1; : : : ; n 1, let Mi 2 Mr be so that Mi \ !1 = i and let gr ( i ) = xi . Also, for i = 0; : : : ; n 1 ,let gi = gr \ Mi and Hi = Hr \ Mi and let gn = gr and Hn = Hr . Just exactly as in the case n = 1, but using Mn 1 in place of M = M0, we obtain that ~ s:t: g = g_ hmax(dom(g0 )); xi&Hn 1 H g] 2 F : cl fx : 9hg; H i 2 D De ne, for i = 0; : : : ; n 1 : ~ n (i+1) = fhg; H i : cl Zn (i+1) hg; H i] 2 Fg; D where for any hg; H i we let (1) Zn (i+1) hg; H i] = ~ n i s:t: g0 = g_ hmax(dom(g0 )); xi and H H 0 g : fx : 9hg0 ; H 0i 2 D ~ n i 2 M0 for all i = 0; : : : ; n, where D ~n = D ~ . Furthermore Note that D ~ n 1. Assume that i < n and that we have noted above that hgn 1; Hn 1i 2 D ~ n i . Now since xn (i+1) 2 Zn (i+1) hgn (i+1); Hn (i+1)i] and hgn i; Hn ii 2 D Zn (i+1) 2 Mn (i+1) , we again obtain that clZn (i+1) 2 F :

20

ALAN DOW

~ 0 2 M0 . Therefore hg0 ; H0i 2 D 0 ; : : : ; Hn 0 1 all in M0 We may now pick, h 0 ; y0i; : : : ; h n 1; yn 1i and H1 , so that fy0 ; : : : ; yn 1g U . These are picked recursively so that for each i = n 1; : : : ; 0 , if 0 = g0 _ gn i n (i+1) h n (i+1) ; yn (i+1)i ; 0 i H0 0 0 ~ n i . To carry out the inductive step then Hn n (i+1) and hgn i; Hn ii 2 D 0 0 0 i ; Hn 0 ii] 2 F . ~ we note that since hgn i ; Hn ii 2 Dn i , we have that clZn i+1 hgn 0 0 Therefore U \ Zn i+1 hgn i; Hn ii] 6= ; and we may choose yn (i+1) 2 M0 \ U \ 0 i ; Hn 0 ii]. Then by elementarity we can choose n i+1 and Hn i+1 in Zn i+1 hgn M as required above.
vi. Large Cardinals and Reflection Axioms

We have seen lots of examples where we had a space hX; T i 2 H ( ) and a property or formula '(v1 ; v2; v3) so that when we took M H ( ) and a parameter A 2 M , we had H ( ) j= '(X; T ; A) (hence M j= '(X; T ; A) ) but H ( ) j= :'(X \ M; T \ M; A \ M ) : In fact the whole point of re ection is to nd conditions on M which are su cient to guarantee that ' does re ect, as opposed to the situation described above. If a cardinal is supercompact and hX; T i; A 2 H ( ) and '(v1 ; v2; v3) is any formula such that H ( ) j= '(X; T ; A) then there is an M H ( ) such that jM j < & fX; T ; Ag 2 M & H ( ) j= '(X \ M; T \ M; A \ M ) (see KaMa]). When we combine this with forcing we get re ection results at \small" cardinals which need large cardinals. To get the most out of this, one would want to master the techniques described in such articles as De1], KaMa] and DTW]. We shall just take the results after the fact as Axioms. PFA is, of course, the Proper Forcing Axiom: Given a proper poset P and a family fD : < !1g of dense open subsets of P , there is a lter G P such that G \ D 6= ; for all < !1. Fleissner has an axiom called Axiom R: If S X ]! is stationary and C X ]<!2 is t.u.b. then 9Y 2 C such that S \ Y ]! is stationary in Y ]! . The set C X ]<!2 is said to be t.u.b. if it is unbounded and if the union of every chain of length !1 from C is again a member of C . Axiom R is a speci c case of a scheme (see DTW] for more details). Roughly speaking, if P is a nice class of forcing notions, then we could have Axiom P : If '(v1 ; v2; v3) is a (local + structural) property which is preserved by forcings from P and if (a; b; c) 2 H ( ) is such that '(a; b; c) holds | then 9Y 2 H ( )]< such that '(a \ Y; b \ Y; c \ Y ) holds and Y \ 2 . For example, for axiom R, take X; S; C 2 H ( ) and P is the class of proper posets of cardinality < = !2. Then '(X; S; C; ) \S is stationary in X ]! , C is unbounded in X ]< and closed

ELEMENTARY SUBMODELS

21

under unions of chains of length !1 ". Now note that proper forcing preserves stationarity { hence '. PFA+ is what you get when you combine PFA and axiom R. For what it's worth, the author nds it easiest to apply PFA by recalling how the consistency of PFA+ is proven. That is, the model in which PFA holds is obtained by forcing with an iteration of length of proper posets. When you are considering a space hX; T i in the extension you know, from the fact that is a large cardinal, that this space and any of its properties will re ect to an inner model { but there will be more forcing to be done. But now the di erence between PFA and the above axiom scheme is that you get to choose the next forcing in the iteration. The idea then is to choose the next forcing so that the iteration of it with any other proper poset will preserve the properties of interest. Other axioms which are frequently used (but not as axioms) are: \the Cohen forcing Axiom" = Axiom Cohen2! ; \Mitchell forcing Axiom " = Axiom (Cohen !1-closed)!2 ; where Cohen !1closed denotes the class of posets which are of the form Fn(!2; 2) P and P is forced to be a countably closed poset by the Cohen posets. \Levy forcing Axiom" = Axiom !1 -closed!2 . Using these axioms together with a judicious choice of ' we can obtain !1 -sized , !-covering elementary submodels of some H ( ) together with some instances of 1 1 -re ection. For example as we promised in xIV . Proposition 6.1: Axiom R ) (X; metriz ) < !2 for X in the class of locally@1 spaces. Hence, in particular, Axiom R implies there are no E -sets. and 4.2, we may assume that = jX j = X is a regular cardinal. Let B be a base for X consisting of open sets of cardinality at most @1 . For our application of Axiom R we de ne S = fs 2 X ]! j s n sup(s) 6= ;g : Next, x a su ciently large cardinal and let M denote the set of !1 -sized, !covering elementary submodels of the structure (H ( ); X; B; 2). Let C = fY 2 X ]!1 : (9 M 2 M)(M \ X = Y )g : Since X is rst-countable and locally-@1, it follows from an easy generalization of 1.4 that C is a t.u.b. subset of X ]!1 . Before we show that S is stationary in X ]! , let us suppose that it is and show how to deduce the result from Axiom R. By Axiom R, we may choose Y 2 C such that S \ Y ]! is stationary in Y ]! . Let M 2 M be such that M \ X = Y . Since M is !1-covering we may choose a continuous elementary 2-chain fM : 2 !1g whose union is M . Choose a large 0 and a continuous 2-chain fN : 2 !1 g of elementary submodels of H ( 0 ) so that fM : 2 !1g 2 N0 . Since S \ Y ]! is stationary, we may choose a 2 !1 such that M \ Y = N \ Y 2 S . By the de nition of S , there is an x 2 Y \ N sup(Y \ N ). We rst observe that, by elementarity, there is such an x in Y . Indeed, rst notice that, since M and X are in M +1 and M \ X = M \ Y = N \ Y , it follows that N +1 j= (9x) (x 2 Y \ M sup(Y \ M )). Finally we use 2.3 again toS contradict that Y is supposed to be metrizable. For this we will show that x 2 = (B \ N ), where x is chosen as above. For each B 2 B \ N , the element sup(B ) is a member of N , hence B sup(N \ X ).
Proof. Let X be a locally-@1 space and assume that X is not metrizable. By 4.1

22

ALAN DOW

Therefore it su ces to show that sup(N \ X ) sup(M \ X ). To see this observe that for each 2 X \ N , sup(f < !1 : fsup(M \ X ) )g) 2 N . Now to prove that S is stationary. Suppose that A is a closed and unbounded subset of X ]! such that A \ S = ;. Let fM j 2 g be a continuous elementary 2-chain of elementary submodels of H ( ) such that, for each 2 , !1 fX; B; A; S g M , jM j < , and M \ 2 . As in the proof of 4.2, each X \ M = \ M is an open metrizable subspace of X . Since X is not metrizable, there is an such that X \ M is not closed. Since X is rst-countable, we may choose a countable set s X \ M such that s n M 6= ;. Therefore s n sup(M \ ) 6= ;. Now let N be a countable elementary submodel of M which contains s fAg; we claim that a = N \ X 2 A \ S . First a 2 S since a s and sup(a) sup(M ). Since N j= (A is an unbounded subset of X ]! ), it follows that there is a countable chain C N \ A such that a = C - hence a 2 A. This contradicts that A was chosen to miss S

Proposition 6.2: It follows from the Mitchell Axiom (and the Levy Axiom) that a space with countable tightness and uncountable character has a !1-sized subspace with uncountable character. That is, (X; !1 ) !1 for any X with countable tightness.
Proof. Let hX; T i have countable tightness and assume that x 2 X has uncountable character. Suppose further that the character of every countable subspace of X is countable (for the Levy Axiom we must assume that every @1 -sized subspace has countable character). Choose a regular cardinal large enough to contain the power set of the power set of X . De ne the formula ' so that 8 t(x; hX; T i) = ! and (x; hX; T i) > ! > > >

Now we show that it is consistent that 3.4 can be improved.

'(x; X; T ; H ( )) i

> < > > > > :

Now we must check that forcing by Cohen !1 -closed preserves that ' holds. Lemma 5.6 (and 5.3 for Levy) proves that countable tightness is preserved. The second line in the de nition of ' is also preserved by any proper forcing but it deserves more discussion. At rst glance it seems a total triviality | but the important point is that we are talking about the set H = H ( ) as opposed to the de ned notion. All the second line is really saying is that \H has the !-covering property " and we are simply asserting that this is preserved by proper forcing. Of course if we had put a = a ~ in line two { this would not have been preserved by any forcing which adds a real. Since H has the !-covering property, uncountable character is preserved. Now either of the above Axioms gives us a set M with jM j = !1 (which we may as well assume is a subset of H ( ) ) so that '(x; X \ M; T \ M; M ) holds: Therefore, in the subset X 0 = X \ M and with respect to the topology induced by T 0 = T \ M , the point x has countable tightness and uncountable character.

8a 2 H ( )]! 9a ~ 2 H ( ) so that ja ~j = ! ; a ~ a ; and H ( ) j= (x; hX \ ~ a; T i = ! :

ELEMENTARY SUBMODELS

23

Therefore to nish the proof we wish to show, just as we were doing in x III , that T 0 induces the subspace topology on X 0 { at least at x. So let x 2 U 2 T and assume that x 2 clT X 0 n U ]. Since we have t(x; hX 0 ; T 0 i) = !, we may choose a countable a X 0 n U so that x 2 clT a. But now M has the !-covering property hence we may choose a countable a ~ 2 M so that a a ~ 2 M . This contradicts that M j= a ~ has countable character with respect to the topology induced by T since M would then contain a base for the subspace topology at x .
0 0

We nish this section with the PFA results on initially !1 -compact spaces of countable tightness. Fremlin and Nyikos proved (i) and (ii) is due to Balogh. In fact Balogh proved that, under PFA, compact spaces of countable tightness are sequential but we do not include this result since it depends on the case in 5.7 which we did not prove.

Theorem 6.3: PFA implies that if X is an initially !1-compact space of countable tightness then : (i) X is compact; (ii) X is sequentially compact; and (iii) X is rst countable at some of its points.
Proof. Let hX; T i be an initially !1 -compact space with countable tightness. Let P

be the usual countably closed collapse of jX j . In the extension obtained by forcing with P , the space hX; T i will not be compact if any of the conditions (i) - (iii) failed to hold. Indeed, if (ii) fails to hold then clearly X contains a closed subspace in which there are no points of rst-countability { so we may as well assume that (iii) fails. For each x 2 X , x a closed G , Fx such that x 2 = Fx . Let \ <! ~ = fg 2 1 X j Fg( ) 6= ;g: P 2dom(g) T ~ is a P ~ -generic branch, then 2!1 FG If G ~ ( ) = ; since each non-empty G subset of X must contain many points. One can now observe that forcing with P will add ~ ; ( or force with P ~ in the rst place, or even that we a generic branch through P may assume without loss of generality that each non-empty G subset of X has the ~ = P ). same cardinality as X hence P In the extension hX; T i is still countably compact and contains no free sequences by 5.7 . Now use 5.8 to nd a proper poset Q in the extension so that there is a P Q-name g _ so that 1 j`g _ is a homeomorphism from !1 into hX; T i : _ ;U _ g : 2 !1g such that Therefore there are also P Q-names ffW

_ ; U g T and (2) 1 j`P Q 8 2 !1 fW _ W _ V _ and V _ \g g _ ( 0; ]) W _ (( ; !1)) = ; : Finally, we de ne D for 2 !1 to be D = fp 2 P Q : 9x; W; U such that _ = W; and U _ = U g. Since the above statements are forced by 1, p j`g _ ( ) = x; W it follows that D is dense for each 2 !1. Use PFA to nd a lter G P which

24

ALAN DOW

meets each D . Pick, for each 2 !1 , x ; W ; U ; and p so that p 2 G \ D and _ = W ; and U _ =U : p j`g _( ) = x ; W Since G is a lter, it follows that for < ; x 2 W and that x 2 = V . Therefore (back in V ) fx : 2 !1g is a free sequence { since we have the same base for the topology in both models, W \ X n V must be empty. This contradicts the fact that X contains no free sequences (in V ) .

Remark. The role of the pair fW ; V g in the above proof is critical. It is not true, in general, that if you introduce a free sequence with proper forcing then you must have had one to begin with. Perhaps the easiest way to see what is going on is to think of the above mentioned \author's-view" of PFA. When you meet !1-many dense sets from the poset P Q, you are really forcing over some inner model. We can think of this forcing as introducing a sequence which is \free with respect to the inner model space". However there are still points to be added to the space which can destroy that freedom. Also there are still neighbourhoods to be added of the points you do have and this is why we do not, and can not, assert that we get a copy of !1 in X .
vii Submodels closed under !-sequences and Forcing

In this last section we will prove a few results that show that the techniques involved when using large cardinals can be used even without the large cardinals. All that is going on in the results of the previous section is that a forcing statement is rst re ected, then the forcing is factored and nally a preservation result is proven. When countable objects seem to determine all the re ection that you need then it is possible that a large cardinal is not needed. It may su ce to re ect the forcing statement (as in the above outline) by simply taking an elementary submodel closed under !-sequences. The more di cult arguments (e.g. those using PFA) may require the assumption of }(!2) because it sometimes depends on the order in which you iterate your posets. If the forcing is simply an iteration of the same poset then you probably just need to assume CH in the ground model as we shall demonstrate below with Cohen forcing. Frequently these results are proven using the -system lemma and other combinatorics . _ : < !2g , fB _ : < !2g and The general procedure is to let, say, fA _ : < !2g be Fn(!2; 2)-names of subsets of !2 . Let A = fA _ : < !2 g and fC similarly de ne B and C . Let M H (!3 ) be such that fA; B; Cg 2 M ; M ! M ; and M \ !2 = < !2 : _ , of a subset of !2 can be assumed to be a Recall that a Fn(!2; 2)-name, say A _ i ( ; p) 2 A _. subset of !2 Fn(!2; 2) where p j` 2 A _ _ _ Now we let A = A \ Fn( ; 2) for each < . Similarly de ne B ! _ . Using the facts that M H (!3) and M M one can easily prove that and C _ =A _ \ and many other re ection results of the form 1 j`Fn( ;2)A _ ;::: ;C : _ ; : : : ; C , 1 j`Fn(!2 ;2)' A 1 j`Fn( ;2)' A

ELEMENTARY SUBMODELS

25

The nal and crucial step after having obtained the validity of the appropriate forcing re ection is to prove that further Cohen forcing preserves the property. Let us begin with a well-know result of Kunen's.

Proposition 7.1: In the model obtained by adding !2-Cohen reals to a model of CH, there are no !2 -chains in P (!) mod fin.
_ j < !2g P (!) and A _ _ for < < !2: 1 j`Fn(!2 ;2)A = fA A _ : 2 !2 g so that Also x a name B for fB _ n B j < ! for all < !2 g: 1 j`B = fB 2 P (!) j jA Let A ; B 2 M H (!3 ) with M ! M; jM j = !1 and M \ !2 = . For < , _ and B _ are Fn( ; 2)-names. Now we may assume that, in fact, A _ : < !2 g = fB j (8 < !2) jA _ n B j < !g: M j= 1 j`fB _ is such that 1 j`Fn( ;2)(8 2 ) jA _ nA _ j < !: Since Fn( ; 2) is ccc and Suppose A _ 2 M so that 1 j`Fn( ;2)A _ =B _ . Therefore, ! is countable, there is a B _ =B _ : 1 j`9 < such that B This application of \!-absoluteness" has shown that if G is Fn(!2; 2)-generic and if G = G \ Fn( ; 2), then _ ; G ) j < g = fB 2 P (!) j (8 < )val(A _ ; G ) B g: V G ] j= fval(B _ ; G) for < !2 . By assumption, jA n A j < ! Now in V G], let A = val(A for all < and jA n B j < ! for all < . Now refer to 5.1 and let I = fB ! : B 2 V G ] and B A g . Since I is countably generated and cf ( ) = !1 , there is an I 2 I so that A I for co nally many 2 . Therefore, A I for all 2 . But then, by the above, there is a 2 such that I = B . But this implies that A = A +1 (a contradiction) since B = I A A +1 B This technique is also useful in proving Malykin's interesting new result. Van Douwen and van Mill have shown that it is consistent that (e.g. under PFA) ! fxg is C -embedded in ! for any point x 2 ! . Malykin has shown that this is also true in the Cohen model. I feel that this result demonstrates that there are still interesting consistency results to be obtained in the Cohen model. _ j < !2 g , are Fn(!2; 2)-names such that Proof. Suppose A; fA

Proposition 7.2: If G is Fn(!2; 2)-generic over V , a model of CH, then, in V G], ! -fxg is C -embedded for each x 2 ! .

26

ALAN DOW

are such that x 2 f 0; r0] \ f r1; 1]. It is well-known that R0 = f 0; r0] and similarly R1 are regular closed subsets of ! . Let fA g <!2 and fB g <!2 be the subsets of ! whose remainders are contained in R0 and R1 respectively. Fix Fn(!2; 2)-names for the A 's and the B 's and nd < !2 just as in 7.1 (you would also want to ensure that x was in M ). T Using 5.1 and the fact that ! is an F-space one can show that R0 \ R1 fC : C 2 x \ V G ]g. Indeed, suppose that D ! is such that D \ X 6= ; for all X 2 x \ V G ] and that D \ R0 = ;. By 5.1, there are fXn : n 2 !g x \ V G ] which generate the lter fY 2 V G ] : D Y g. This set fXn : n 2 !g need not be in V G ] in general but since x is a lter T we can enlarge the set fXng so that we may assume that it is in V G ]. Let Z = fXn : n 2 !g and note that x 2 Z \ R0 and that Z \ R0 is again regular closed. But now Z 2 V G ], hence we may choose an < so that A 2 V G ] and A Z \ R0. This contradicts that there should be an n so that Xn ! n A . To nish the proof then we just have to note that x \ V G ] does not generate x. We nish with a new proof of a result from DTW]. The original proof of this (and the PMEA analogue) involved rather more di cult lter combinatorics.

Sketch of Proof. Assume that f ! -fxg ! 0; 1] is continuous and that r0 < r1

Proposition 7.3: If G is Fn(!2; 2)-generic over V , a model of CH, then, in V G], a rst countable space of weight !1 is metrizable if each of its @1 -sized subspaces are metrizable.
_ : 2 !1g be Fn(!2; 2)-names of subsets of !2 so that 1 j`h!2 ; fB _ j Proof. Let fB < !1gi is a rst countable space in which each subspace of size !1 is metrizable. _ g <!1 2 M . Let Let M H (!3 ) be so that M ! M , jM j = !1 and fB _ = M \ !2 ; G = G \ Fn( ; 2) and let fB g <!1 be as above. Then V G ] j= _ g < i has a _ g <!1 i is a rst countable space. By 5.4, 1 j`Fn( ;2)h ; fB h ; fB _ so that 1 j`U _ !1 !1 ! and so that U _ -discrete base. Fix a Fn( ; 2)-name U th \codes" a -discrete base for . That is, the ( ; n) member of the base will be _ : ( ; ; n) 2 Ug _ . We will show that the collection whose ( ; n)th the union of fB _ _ will form a -discrete base for the member is the union of fB : ( ; ; n) 2 Ug whole space. _ were a member of M but there is no reason to We would be done if the name U suppose that this would be so. However, the trick is to isolate, for each remaining _ which will do the job. This countable piece x 2 !2 a countable piece of the name U will be in M which will allow us to play the !-absoluteness game. _ : < 0g contains a base Let x 2 !2 and let 0 2 !1 be such that 1 j`fB at x. Let N be a countable elementary submodel of H (!3) which contains the set _ : < 0g; U _ ; M g. Now let = N \ !1 and let U _N = U _ \ N . Since M is fx; fB _ closed under !-sequences, UN 2 M . _ ) denote the formula (with parameter fB _ j 2 !1g): Let '( ; U _ \1 j` If y 2 !2 is such that fB j 2 g contains a base for y, then, (i) for each n 2 !, y has a neighbouhood meeting at most one member of the family _ j ( ; ; n) 2 U _ g j 2 !1g, and f fB

ELEMENTARY SUBMODELS

27

_ , there are 2 ! and n 2 ! such that (ii) for each 2 , such that y 2 B _ _ y 2 fB j ( ; ; n) 2 U g B ." Now we observe that : _ N ): M j= '( 0 ; U Therefore, _ N ): H (!3) j= '( 0 ; U _N = U _ \ N , we have: But now, since U _ ): N j= '( 0; U And, nally, since N H (!3 ), we obtain that: _ ): H (!3 ) j= '( 0; U _ codes This completes the proof since it shows that, at least with respect to x, U a -discrete base.
References
1] A. V. Arhangel'skii, On the cardinality of bicompacta satisfying the rst axiom of countability , Soviet Math. Dokl. ,10 ,1969 ,951-955 2] ibid., Structure and classi cation of Topological Spaces and Cardinal Invariants , Uspehi Mat. Nauk. ,33 ,23-84 ,1978 3] Zoltan Balogh, On the structure of compact spaces of countable tightness , to appear Proc. AMS 4] Keith J. Devlin, The Yorkshireman's guide to the Proper Forcing Axiom , Proc. 1978 Cambridge Summer School in Set Theory ,1978 5] ibid., The Axiom of Constructibility ,Springer Verlag ,1984 6] A. Dow, Removing large cardinals from the Moore-Mrowka solution , preprint 7] A. Dow, I. Juhasz and W. A. R. Weiss, On increasing chains of rst countable spaces , to appear in Israel J. Math. 8] A. Dow, F. Tall and W. A. R. Weiss, New proofs of the Normal Moore space Conjecture , to appear in two parts in Top. Appl. 9] W.G. Fleissner, Left separated spaces with point-countable bases , Trans. AMS ,294 ,1986 ,665-677 10] David Fremlin, Perfect pre-images of !1 , to appear 11] David Fremlin and Peter Nyikos, Initially !1 -compact spaces are compact under PFA , Private Communication 12] Istvan Juhasz, Cardinal Functions in Topology: Ten Years Later ,Math. Center Tracts ,Amsterdam 13] A. Kanamori and M. Magidor, The evolution of large cardinal axioms in set theory , Proc. Conf. on Higher Set Theory ,Lect. Notes Math. ,669 ,1978 ,99-275 14] Kenneth Kunen, Set Theory: An Introduction to independence proofs ,North Holland ,1978 15] M. Magidor, On the Role of Supercompact and Extendible Cardinals in Logic , Israel J. Math. ,10 ,1971 ,147-171
Department of Mathematics, York University, North York, Ontario, Canada M3J 1P3