HEREDITARILY LINDELOF SPACES OF SINGULAR DENSITY

ISTVAN JUHASZ AND SAHARON SHELAH Abstract. A cardinal is called -inaccessible if for all < we have < . We show that for every -inaccessible cardinal there is a CCC (hence cardinality and conality preserving) forcing that adds a hereditarily Lindelf regular space of density . This o extends an analogous earlier result of ours that only worked for regular .

modified:2007-03-19

In [1] we have shown that for any cardinal a natural CCC forcing notion adds a hereditarily Lindelf 0-dimensional Hausdor topology o on that makes the resulting space X left-separated in its natural well-ordering. It was also shown there that the density d(X ) = cf(), hence if is regular then d(X ) = . The aim of this paper is to show that a suitable extension of the construction given in [1] enables us to generalize this to many singular cardinals as well. Note that the existence of an L-space, that we now know is provable in ZFC (see [3]), is equivalent to the existence of a hereditarily Lindelf regular space of density 1 . Since the cardinality of a hereditarily o Lindelf T2 space is at most continuum, just in ZFC we cannot replace o in this 1 with anything bigger. The following problem however, that is left open by our subsequent result, can be raised naturally. Problem 1. Assume that 1 < c. Does there exist then a hereditarily Lindelf regular space of density ? o After this paper was submitted and accepted, the second author gave a negative answer to this question. This will be included in a forthcoming paper that is listed as Sh:F821 in the so called F-list of still unpublished works of his and will appear under the title Consistent partition relations below the power set. Before describing our new construction, let us recall that the one given in [1] is based on simultaneously and generically splitting into
2000 Mathematics Subject Classication. 54A25, 03E35. Key words and phrases. hereditarily Lindelf space, density of a space, singular o cardinal, forcing. The rst author was supported by OTKA grant no. 61600.
1

899

revision:2007-03-19

I. JUHASZ AND S. SHELAH

two the complements \ for all proper initial segments of . The novelty in the construction to be given is that we shall perform the same simultaneous splitting for the complements of the members of a family A of subsets of that is, at least when is singular, much larger than the family of its proper initial segments (that is just if we are considering von Neumann ordinals). The following denition serves to describe the properties of such a family of subsets of . Denition 2. Let be an innite cardinal. A family A of proper subsets of is said to be good over if it satises properties (i)-(iii) below: (i) A that is, all proper initial segments of belong to A; (ii) for every subset S with |S| < there is A A with S A; (iii) for every subset S with |S| = 1 there is T [S]1 such that if A A then either |A T | or T A. If is regular then A = , the family of all proper initial segments of , is a good family over . Indeed, (i) and (ii) are obviously valid and if S []1 then any subset T of S of order type 1 satises (iii). If, however, is singular then this A denitely does not satisfy condition (ii). Actually, we do not know if it is provable in ZFC that for any (singular) cardinal there is a good family over . But we know that they do exist if is -inaccessible, that is < holds whenever < .
modified:2007-03-19

Theorem 3. If is an -inaccessible cardinal then there exists a good family A []< over . Proof. It is well-known that there is a map G : [] with the property that for every a [] we have G [a] = . In other words: we may color the innite subsets of with countably many colors so that on the subsets of any innite set all the colors are picked up. Such a coloring may be constructed by a simple transnite recursion. Next we x a maximal almost disjoint family F of subsets of order type of our underlying set and then we transfer the coloring G to each member F of F . More precisely, this means that for every F F we x a map GF : [F ] F such that GF [a] = F whenever a [F ] . Then we t together these colorings GF to obtain a coloring H : [] of all countable subsets of as follows: For any S [] we set H(S) = GF (S) if there is an F F with S F and H(S) = 0 otherwise. The coloring H is well-dened because, as F is almost disjoint, for every S [] there is at most one F F with S F. Now, a set C is called H-closed if for every S [C] we have H(S) C. Clearly, for every set A there is a smallest H-closed set

899

revision:2007-03-19

 LINDELOF

including A that will be denoted by clH (A) and is called the H-closure of A. Let us set A+ = A H [A] for any A . It is obvious that then we have clH (A) = A ,
<1

where the sets A are dened by the following transnite recursion: A0 = A, A+1 = (A )+ , and A = < A for limit. Since H [A] {F F : |F A| = } {0} ,

it is also obvious that we have |A+ | |A| for all A and consequently |clH (A)| |A| as well. In particular, |A| < implies |clH (A)| < because is inaccessible. Now we claim that the family A of all H-closed sets of cardinality less than is good over . Indeed, rst notice that because each F F has order type , for every set S [F ] we have H(S) = GF (S) < sup F = sup S, implying that every initial segment of is H-closed and so A satises condition (i) of denition 2. Condition (ii) is satised trivially. To see (iii) we rst show that there is no innite strictly descending sequence of H-closed subsets of , or in other words: the family of Hclosed sets is well-founded with respect to inclusion. Assume, reasoning indirectly, that {Cn : n < } is a strictly decreasing sequence of Hclosed sets and for each n < we have n Cn \ Cn+1 . By the maximality of F then there is some F F such that the set S = F {n : n < } is innite. Then, for any k < , the set S Ck is also innite and consequently we have H [S Ck ] = GF [S Ck ] = F Ck because Ck is H-closed. But for any m < such that m S this would imply m F Cm+1 , which is clearly a contradiction. Now let S with |S| = . Our previous result clearly implies that there is a set T [S] such that we have clH (U ) = clH (T ) whenever U T with |U | = . In other words, this means that for every H-closed set C we have either |C T | < or T C. In particular, for = 1 this shows that our family A satises condition (iii) of denition 2 as well, hence it is indeed good over .

899

revision:2007-03-19

modified:2007-03-19

I. JUHASZ AND S. SHELAH

Problem 4. Is it provable in ZFC that for every (singular) cardinal there is a good family over ? Next we present our main result that, in view of theorem 3, immediately implies the consistency of the existence of hereditarily Lindelf o regular spaces of density practically for any singular cardinal . (Of course, this has to be in a model in which c.) We shall follow [2] in our notation and terminology concerning forcing. Theorem 5. Let A be a good family over . Then there is a complete (hence CCC) subforcing Q of the Cohen forcing F n(A , 2) such that in the generic extension V Q there is a hereditarily Lindelf 0o dimensional Hausdor topology on that has density . If we also have A []< (as in theorem 3) then every subset of of size < is even -nowhere dense. Proof. We start by dening the the subforcing Q of F n(A , 2): Q consists of those p F n(A, 2) for which A, dom p with A implies p(A, ) = 0 and A, A dom p implies p(A, A ) = 1, where A = min( \ A). It is straight-forward to check that Q is a complete suborder of F n(A , 2). For any condition p Q and any set A A we dene
p UA = { : p(A, ) = 1},
modified:2007-03-19

and if G Q is generic then, in V [G], we set UA =

p {UA : p G}.

899

revision:2007-03-19

0 1 Next, let UA = UA and UA = \ A and be the topology on i generated by the sets {UA : i < 2, A A}. Note that then the family (A) B = {B : F n(A, 2)} is a base for , where B = Adom UA . It is clear from the denition that each B is clopen, hence is 0dimensional. Now, if < < then we have A by (i) and hence 0 1 U while = U , which shows that is also Hausdor. It is also immediate from (ii) that no set S []< is -dense, hence the space , has density . Indeed, if S A A then we have 1 1 S UA = , while UA = . Thus it only remains for us to prove that the topology is hereditarily Lindelf. o Assume, reasoning indirectly, that some condition p Q forces that is not hereditarily Lindelf, i. e. there is a right separated 1 o sequence in . More precisely, this means that there are Q-names s and e such that p forces s : 1 , e : 1 F n(A, 2), s() Be() , and s() Be() whenever < < . Then, in the ground model V , /

 LINDELOF

for each < 1 we may pick a condition p p, an ordinal < , and a nite function F n(A, 2) such that p s() = e() = . Since Q is a complete suborder of F n(A , 2) it has property K, hence we may assume without any loss of generality that the conditions p are pairwise compatible. By extending the conditions p , if necessary, we may assume that dom p = I a with I [A]< and a []< , moreover dom I and a whenever < 1 . With an appropriate thinning out (and re-indexing) we can achieve that if < < 1 then a {A : A I }. / Using standard counting and delta-system arguments, we may assume that each has the same size n < , moreover the sets dom = {Ai, : i < n} [A]n form a delta-system, so that for some m < n we have Ai, = Ai if i < m for all < 1 , and the families {Am, , ..., An1, } are pairwise disjoint. We may also assume that for every i < n there is a xed value li < 2 such that (Ai ) = li for all < 1 . With a further thinning out we may achieve to have dom I = {Ai : i < m} whenever < < 1 . Finally, by property (iii) of the good family A, we may also assume that the set T = { : < 1 } []1 satises either |A T | or T A whenever A A. Now, after all this thinning out, we claim that there is a countable ordinal > 0 such that, for every i < n, if Ai,0 then li = 0. Indeed, arguing indirectly, assume that for every 0 < < 1 there is an i < n with Ai ,0 and li = 1. Then there is a xed j < n such that the set { : i = j} is uncountable and lj = 1. But the rst part 0 implies |Aj,0 T | = 1 , hence 0 T Aj,0 UAj,0 that would imply 0 (Aj,0 ) = lj = 0, a contradiction. So, let us choose > 0 as in our above claim. We then dene a nite function q F n(A , 2) by setting q p0 p , dom q = dom p0 dom p { Ai,0 , : m i < n}, and nally q(Ai,0 , ) = li for all m i < n. We have a0 , and also Ai,0 I for m i < n / / by our construction, hence this denition of q is correct. Moreover, by
modified:2007-03-19

899

revision:2007-03-19

I. JUHASZ AND S. SHELAH

the above claim if Ai,0 then li = 0 and if Ai,0 then = Ai,0 , / consequently we actually have q Q. Let us observe, however, that we have q(Ai,0 , ) = li for all i < n. Indeed, if i < m then this holds because p (Ai,0 , ) = p (Ai , ) = li . But this implies that q B0 and hence q s() Be(0) that is clearly a contradiction because q extends p. Now assume that we also have A []< (in V ). Since Q is CCC, every subset of in V Q is covered by a ground model set of the same size, hence it suces to show that any ground model member Y of []< is -nowhere dense. To see this, we rst note that it follows from a straight-forward density argument that for every F n(A, 2) we have |B | = . (Actually, this only uses the assumption that | \ A0 | = for every A0 [A]< which is weaker than A []< .) Next, consider any set Y []< V and a xed F n(A, 2). Since A satises condition (ii) of denition 2, we may clearly nd an A A such that Y A and A dom . Let = { A, 1 }, / 1 then B = B UA is a non-empty open subset of B that is clearly disjoint from A and hence from Y as well. This shows that Y is indeed -nowhere dense. For a singular cardinal of conality the results of [1] did imply the existence of hereditarily Lindelf regular spaces of density , by taking o the topological sum of those of density n with n regular and = n< n . It should be emphasized, however, that the spaces obtained in this way clearly do not have the stronger property we obtained in theorem 5 that all subsets of size less than are nowhere dense. So, we do have here something new even in the case of singular cardinals of conality . Finally, we would like to point out that the forcing construction given in [1] may be considered as a particular case of that in theorem 5, where the good family A over happens to be equal to the family of all proper initial segments of .

revision:2007-03-19

modified:2007-03-19

References
[1] I. Juhsz and S. Shelah, Generic left separated spaces and calibers, Top. Appl. a 132 (2003), 103108. [2] K. Kunen, Set Theory, North-Holland, Amsterdam, 1980 [3] J. T. Moore, A solution to the L space problem, J. Amer. Math. Soc. 19 (2006), 717736.

899

 LINDELOF

Alfrd Rnyi Institute of Mathematics e e E-mail address: juhasz@renyi.hu Einstein Institute of Mathematics, The Hebrew University of Jerusalem E-mail address: shelah@math.huji.ac.il

899

revision:2007-03-19

modified:2007-03-19