Sei sulla pagina 1di 8

COVERING NUMBERS ASSOCIATED WITH TREES BRANCHING INTO A COUNTABLY GENERATED SET OF POSSIBILITIES

SAHARON SHELAH

Contents 1. Introduction 2. Equivalences 3. Trees of Countable Structures References 1 2 3 8

1. Introduction This paper is concerned with certain generalizations of meagreness and their combinatorial equivalents. The simplest example, and the one which motivated further study in this area, comes about by considering the following denition:
modified:1998-07-04

Denition 1. For any A R a set X R will be said to be A-nowhere dense1 if and only if for every q A there exists and integer k Z such that the interval whose endpoints are q and q + 1/k is disjoint from X. A set which is the union of countably many A-nowhere dense sets will be called A-very meagre. The notion of an A-nowhere dense set for various subsets A of the reals may prove to be of interest in its own right, but this paper will be concerned exclusively with the special case A = Q. Notice that rational perfect sets introduced by Miller in [1] form a subset of the Q-nowhere dense sets since the closure of a set is rational perfect if it is perfect and disjoint from the rationals. On the other hand, a set is Q-nowhere dense if its 2-sided-closure is disjoint from the rationals where the 2-sided-closure of a set refers to all those reals which are limits of both decreasing and increasing sequences from the set.
1991 Mathematics Subject Classication. Primary 03E35; Secondary 28A05. The author would like to thank Juris Steprns for writing up this paper. This paper a is number 660 in the authors personal listing. 1This notation corrects the terminology of [4] which called a special case of this notion almost nowhere dense in spite of the fact that almost nowhere dense sets are nowhere dense rather than the converse.
1

660

revision:1997-11-06

S. SHELAH

In [4] the least number of Q-meagre sets required to cover the real line is examined and is denoted by d1 . It is shown in [3] that there is a continuous function H rst constructed by Lebesgue such that the least number of smooth functions into which H can be decomposed is equal to d1 . This paper will further study d1 and some of its generalizations. As well, an equivalence will be established between Q-meagreness and certain combinatorial properties of trees. This will lead to new cardinal invariants and various independence results about these will then be established. 2. Equivalences Denition 2. A set X Z is small if for each n and : n Z there is some k Z such that either {f (n) : f X and f } {m Z : m < k} or {f (n) : f X and f } {m Z : m > k} It will be shown that the least number of very meagre sets required to cover R is equal to least number of small sets required to cover Z. This is an immediate consequence of the following lemma. The idea is to modify one of the standard2 constructions of a homeomorphism between and R \ Q. In this construction is identied with Z and nite sequences : m Z are mapped to open intervals I() so that if then I() I( ) nZ I( n) = I() the left endpoint of I( n) is the right endpoint of I( (n 1)). This construction will be modied by, essentially, mapping sequences which end with a negative integer to their counterpart above the neighbour to the right of their parent sequence. The details are explained in the following. Lemma 1. Let X Z be the set of all sequences eventually equal to 1 together with the constant sequence 0. There is a bijection F : Z\X R\Q such that A R \ Q is very nowhere dense if and only if F 1 A is small. Proof. To begin, let {qn }n enumerate Q so that each rational occurs in nitely often. Let Z denote the set of nite sequences of integers in other words, the set of functions from an integer to Z. For Z dene () = (n) where n is the greatest integer in the domain of . Let be the partial ordering of Z dened by if there exists some least integer n such that (n) = (n) and 0 (n) < (n) there exists some least integer n such that (n) = (n) and (n) < (n) < 0
2For example see page 5 of [2].

660

revision:1997-11-06

modified:1998-07-04

COVERING NUMBERS OF TREES

there exists some least integer n such that (n) = (n) and (n) < 0 (n) For Z and n Z dene n Z such that and n have the same domain, (i) = n (i) if i is not the maximal element of the common domain of and n and () + n = ( n ). Moreover, let = (|| 1). Next, construct by induction a mapping f : Z R such that: f is a order preserving mapping whose range is disjoint from Q if () = 1 and is 1 on its domain, or 0, 1, () 0, then f ( 1 ) < f ( i) < f () for each i Z \ {0, 1}, and f ( 1) = f ( 1 ), f ( 0) = f () if () = 0 and is 0 on its domain, or 0, 1, () < 0, then f () < f ( i) < f ( 1 ) for each i Z \ {0, 1}, and f ( 1) = f (), f ( 0) = f ( 1 ) limn f ( n ) = limn f ( n ) Q for every Z 1 |f () f ( )| < k+1 for every and in kZ such that is the immediate successor of with respect to in kZ 1 if 0 then f ( 0) = f () + ||+1 , f ( 1) = f () 1 if 1 then f ( 1) = f () ||+1 , f ( 0) = f () if , kZ are two successive sequences (with respect to ), and both qk and f ( 1) are between f () and f ( ) then lim n f ( n) = limn f ( n) = qk
modified:1998-07-04

For Z dene an interval of reals by I() = [f (), f ( )], where is suitably either successor or predecessor of in || Z, and dene F : Z\X R \ Q by taking the intersection along a branch in other words, F () is the unique element of n I( n). It is easy to check that this mapping has the desired properties.

3. Trees of Countable Structures A cover on a countable set X is a countable subset B P(X) such that X B / if B B and b [X]<0 then B b B.

revision:1997-11-06

For a cover B on X dene B = {Y X : (A B)(Y A)} and dene B + = P(X) \ B. Dene a B-tree to be a tree T X such that for each t T the set of successors of t in T belongs to B or, to be more precise, {s(|t|) : s T and t s} B. The notation will be used to denote the set of all functions from a proper, initial segment of to . Finally, dene JB to be the ideal generated by all sets X such that there is some B-tree T such that X T = {f : (n )f n T }. Note that J B is a countably complete ideal.

660

S. SHELAH

The examples of covers with which this paper will be concerned are of the form Bn = {A n : (i n)|{m : (m, i) A}| < 0 } although many other examples are possible. It will be shown that for any integer m > 1 it is consistent that cov(J Bm ) = 2 but cov(JBm+1 ) = 1 . Denition 3. For a cover B on a set X dene P(B) to be the set of all triples (t, F, ) such that: (1) t X is a nite subtree in particular, t is closed under initial segments (2) F : t B (3) [ X]<0 (4) there is a one-to-one function : t which maps onto the maximal nodes of t such that (x) x for all x (5) x(n) F (x n) for every x and for every n such that x nt The ordering on P(B) is coordinatewise containment. Observe that if (t, F, ) and (t , F , ) are in P(B) and t = t , F = F then the two conditions are compatible. If B is a cover on X and C is a cover on Y then dene B C if and only if for every A B + and for every H : A Y there is B A such that B B + such that there is a nite t Y and a mapping F : t C and there is a nite set C Y and B0 B1 B such that B0 B1 B + and: (3.1) ({b, b } [B0 ]2 )( t)({y, y } [F ( )]2 )( y H(b) and y H(b )) (3.2) (n )(|{b B1 : (c C)(H(b) n c)}| < 0 )

modified:1998-07-04

Denition 4. A (J , )-Lusin set for an ideal J on a set X is a set L X such that |L J| < for all J J . Lemma 2. If B and C are covers on X and Y respectively, B C, is a cardinal of uncountable conality and W is a (J B , )-Lusin set then 1 P(C) W is a (JB , )-Lusin set. Proof. If not, then let T be a P(C)-name for B-tree such that 1 {w } T W and w = w if =

revision:1997-11-06

P(C)

and, for each , choose (t , F , ) P(C) deciding the value of w in other words, such that there is y such that (t , F , ) P(C) w = y . Then, using the fact that has uncountable conality, choose t and F such that W = {y : t = t and F = F } has cardinality . It follows W is + also a (JB , )-Lusin set and, hence, that it is in JB . Next, choose m and : m such that S0 = {s X : (w W )( s w)} B +

660

COVERING NUMBERS OF TREES

and choose W W such that for each s S0 there is a unique ws W such that ws (m) = s. Let (s) be the unique ordinal such that w s = y(s) . Let the maximal nodes of t be enumerated by { i }k and let bs be the unique i i=1 i member of (s) such that i bs . Proceed by induction to dene S 0 and i i+1 i+1 i i i S1 for i k so that if S0 and S1 have been chosen then S0 S1 i+1 i+1 i i Si = S0 S1 is chosen so that S0 S1 B + and so that there is a nite ri+1 Y and a mapping Fi+1 : ri+1 C and there is a nite set Li+1 Y such that (3.3) i+1 s ({s, s} [S0 ]2 )( ri+1 )({y, y } [Fi+1 ( )]2 )( y bs and y bi+1 ) i+1 (3.4) (3.5) (3.6) (3.7) (3.8) (3.9)
i+1 (n )(|{s S1 : (c Li+1 )(bs i+1

n c)}| < 0 )

(ri Li )

Fi (i ) = F (i ) (c Li )( max(ri ))( c) (c Li )( ri )( c c(||) F ()) if i+1 and n < | | then (n) Fi+1 ( n)

modified:1998-07-04

This is easily accomplished using the denition of B C. Then dene a new condition (t , F , ) such that t k ri t , F k Fi F and i=1 i=1 k Li . i=1 It suces to show that (t , F , ) P(C) {s Sk : ws T } B because / this would contradict that 1 P(C) T is a B-tree. Therefore suppose that (t , F , ) is a condition extending (t , F , ) such that (t , F , )
P(C)

{s Sk : ws T } B

for some B B. In order to obtain a contradiction, notice that for each i i the set i = {s Sk S0 : ( t )( bs and bs (| |) Fi ( ))} is nite / 0 i i because of (3.3) and the fact that ri t . On the other hand, it is possible to choose N so large that for each i and each c L i the sequence c N does not belong to t . Then for each i the set
i i = {s Sk S1 : (c Li )(bs 1 i

revision:1997-11-06

N c)}

is also nite. Therefore it is possible to choose


k

s Sk \ (B
i=1

(i i )) 0 1

and let = First note that it is easy to extend t to t and F to F so that (t , F , ) satises Conditions 1, 2 and 4 of Denition 3 because [Y ] <0 C. Since Condition 3 is also satised it suces to show that (t , F , ) satises

{bs }k . i i=1

660

S. SHELAH

Condition 5 in Denition 3. This, in turn, follows from consideration of i the two cases. First, if s S0 then s i and so there is some ri / 0 ( ) such that y bs and y t . Hence / and y Fi ( ) = F ( ) = F i there is no contradiction to Condition 5 because any such contradiction i would already have occured in the condition (t (s) , F(s) , (s) ). If s S1 s then there is some c Li such that bi N = c N and c does not violate Condition 5. Moreover the choice of N guarantees that c violates Condition 5 if and only if bs does. But now (t , F , ) obviously extends i (t, F, (s) ) = (t(s) , F(s) , (s) ) which forces that ws T and, hence, that s B and this is a contradiction. Corollary 1. Suppose that B is a cover and that W is a (J B , )-Lusin set. Suppose also that and cof() > 0 and that {C } are covers such that B C for each . If, furthermore, P is the nite support iteration of {P(C )} then 1 P W is a (JB , )-Lusin set. Proof. Proceed by induction on . If is a limit ordinal and the lemma fails then let G P be generic and choose a J B -tree T and {p } G such that p P w T W and such that the w are all distinct. Then use the nite support of the iteration and the fact that the conality of is less than to conclude that there is some such that G P where P is the nite support iteration of {P(C )} contains of the conditions {p } . This contradicts the induction hypothesis because the closure of the corresponding w will form a JB -tree contained in T . At successors, use the induction hypothesis, Lemma 2 and the absoluteness of the relation .
modified:1998-07-04

Lemma 3. If P is the nite support iteration of length of the partial orders P(B) for some cover B and G is P generic over V then, in V [G], there are countably many sets in JB whose union covers V . Proof. Standard. Theorem 1. If 0 > 1 are uncountable regular cardinals and B 0 and B1 are covers such that B0 B1 then it is consistent that cov(JB0 ) = 0 and cov(JB1 ) = 1 Proof. Let V |= GCH & ZFC and let P0 be Cohen forcing for adding 0 Cohen reals and let P1 be the nite support iteration, of length 1 of the partial orders P(B1 ). Let P = P0 P1 and let G be P generic over V Notice rst that, since it is easily veried that each member of any J B is meagre, it follows that the 0 -Lusin set added by P0 is also an (JB0 , 0 )Lusin set. From Corollary 1 it follows that V [G] has a (J B0 , 0 )-Lusin set. Hence cov(JB0 ) 0 . Since the continuum in V [G] is 0 it follows that cov(JB0 ) = 0 in this model. On the other hand, it follows from Lemma 3 that cov(JB1 ) 1 . Observe that the nite support iteration P 1 adds Cohen reals over each intermediate model. Moreover, as it has already been observed that each member of any JB is meagre, it follows no family of size

660

revision:1997-11-06

COVERING NUMBERS OF TREES

less than 1 elements of JB1 can cover all the reals. Hence cov(J B1 ) = 1 in this model.
i Denition 5. Dene Bn to consist of all sets

B(a, k) = ( a) (k (n \ a)) where k and a [n]i .


i It is easy to check that each Bn is a cover on n if i < n. i Lemma 4. If i + 1 j then Bn j Bm .

i Proof. Let A (Bn )+ and suppose that H : A ( m). Without loss of generality it may be assumed that A = ka Ak {k} where a [n]i+1 and each Ak is innite. Let D be the set of all k a such that the range of H Ak {k} is contained in a nite branching tree. By compactness, for each k D there is ck ( m) and B k [Ak {k}]0 such that for all N the set {x B k : H(x) N ck } is nite. For k a \ D there must exist some tk ( m) such that

{z m : (x Ak {k})(tk z H(x))} is innite. It is then possible to choose J(k) m and B k [Ak {k}]0 such that {z m : (x B k )(tk z H(x))} {J(k)} and, if {x, y} [B k ]2 and H(x) = tk x and H(y) = tk y then x = y . Now let B0 = ka\D B k and t = {tk : k a \ D}. For s t let
modified:1998-07-04

F (s) = { {J(k)} : tk = s} and let B1 = kD and C = {ck : k a \ D}. After shrinking B k further to guarantee that if k0 , k1 a \ D, tk0 = tk1 , J(k0 ) = J(k1 ) and x B k0 , y B k1 then H(x)(|tk0 |) = H(y)(|tk0 |), it is routine to check that these j i choices witness the required instance of B n Bm . Lemma 5. If j < n m then cov(JBn ) = cov(JBm ). j j Proof. Let m,n : m n be a bijection which is the identity on (n 1). This induces a bijection m,n : ( m) ( n) which sends members of the ideal JBm to members of the ideal JBn . Hence j j cov(JBn ) cov(JBm ). j j To prove the other inequality, proceed by induction on n j + 1 to show that cov(JBn ) cov(JBj ). Now, let T be a family of no more than j
n+1

Bk

revision:1997-11-06

j cov(JBn ) many Bn -trees such that ( n) = T T T . It follows that j j+1 j+1 1 n+1,n (T ) is the closure of a Bn+1 -tree. Next, note that if T is a Bn+1 -tree then, by identifying the successor nodes of T with the elements of (j +1),

660

S. SHELAH

it is possible to construct a natural bijection T from T to ( (j + 1)) which also induces a bijection T : T ( (j + 1)). Furthermore, if A JBj then T (A) JBj . By the induction hypothesis it is known 1
j+1 n+1

that cov(JBn ) = cov(JBj ) and so it is possible to nd a family S T of no j


j+1

j more than cov(JBn ) many Bj+1 -trees such that ( (j + 1)) = T S T . j Now, for T T and S S let B(T, S) = T S and note that B(T, S) is 1 j contained in the closure of a Bn+1 -tree. Hence {B(T, S) : T T and S S} witnesses that cov(JBn ) cov(JBj ). j
n+1

Corollary 2. If j < n, i < m and j > i. then it is consistent that cov(JBn ) < cov(JBm ). i j Proof. Apply Lemma 4 and Lemma 5 together with Theorem 1. As a nal remark, notice that an immediate consequence of Theorem 1 and Lemma 1 is that, for any pair of uncountable, regular cardinals 0 > 1 , it is consistent that cov(M) = 0 and d1 = 1 or, in other words, the covering number of the meagre ideal is 0 while the covering number of the Q-meagre ideal is 1 . References
[1] A. Miller. Rational perfect set forcing. In D. A. Martin J. Baumgartner and S. Shelah, editors, Axiomatic Set Theory, volume 31 of Contemporary Mathematics, pages 143 159, Providence, 1984. American Mathematical Society. [2] Arnold W. Miller. Descriptive Set Theory and Forcing, volume 4 of Lecture Note in Logic. Springer, Berlin, 1995. [3] J. Steprns. Decomposing with smooth sets. Trans. Amer. Math. Soc., 0:00, 1900. a [4] J. Steprns. A very discontinuous Borel function. J. Symbolic Logic, 58:12681283, a 1993. Institute of Mathematics, Hebrew University

660

revision:1997-11-06

modified:1998-07-04

Potrebbero piacerti anche