TIE-POINTS AND FIXED-POINTS IN N

ALAN DOW AND SAHARON SHELAH

Abstract. A point x is a (bow) tie-point of a space X if X \ {x} can be partitioned into (relatively) clopen sets each with x in its closure. Tie-points have appeared in the construction of non-trivial autohomeomorphisms of N\N (e.g. [10, 8]) and in the recent study of (precisely) 2-to-1 maps on N \ N. In these cases the tie-points have been the unique xed point of an involution on N \ N. This paper is motivated by the search for 2-to-1 maps and obtaining tie-points of strikingly diering characteristics.

1. Introduction A point x is a tie-point of a space X if there are closed sets A, B of X such that {x} = A B and x is an adherent point of each of A and B. We picture (and denote) this as X = A x B where A, B are the closed sets which have a unique common accumulation point x and say that x is a tie-point as witnessed by A, B. Let A x B mean that there is a homeomorphism from A to B with x as a xed point. If X = A x B and A x B, then there is an involution F of X (i.e. F 2 = F ) such that {x} = x(F ). In this case we will say that x is a symmetric tie-point of X. An autohomeomorphism F of N \ N (or N ) is said to be trivial if there is a bijection f between conite subsets of N such that F = f N \ N. If F is a trivial autohomeomorphism, then x(F ) is clopen; so of course N \ N will have no symmetric tie-points in this case if all autohomeomorphisms are trivial. If A and B are arbitrary compact spaces, and if x A and y B are accumulation points, then let A x=y B denote the quotient space of
Date: August 12, 2007. 1991 Mathematics Subject Classication. 03A35. Key words and phrases. automorphism, Stone-Cech, xed points. Research of the rst author was supported by NSF grant No. NSF-. The research of the second author was supported by The Israel Science Foundation founded by the Israel Academy of Sciences and Humanities, and by NSF grant No. NSF- . This is paper number in the second authors personal listing.
1

916

revision:2007-08-12

modified:2007-08-12

A. DOW AND S. SHELAH

A B obtained by identifying x and y and let xy denote the collapsed point. Clearly the point xy is a tie-point of this space. We came to the study of tie-points via the following observation. Proposition 1.1. If x, y are symmetric tie-points of N \ N as witnessed by A, B and A , B respectively, then there is a 2-to-1 mapping from N \ N onto the space A x=y B . The proposition holds more generally if x and y are xed points of involutions F, F respectively. That is, replace A by the quotient space of N \ N obtained by collapsing all sets {z, F (z)} to single points and similary replace B by the quotient space induced by F . It is an open problem to determine if 2-to-1 continuous images of N \ N are homeomorphic to N \ N [5]. It is known to be true if CH [3] or PFA [2] holds. There are many interesting questions that arise naturally when considering the concept of tie-points in N \ N. Given a closed set A N \ N, let IA = {a N : a A} . Given an ideal I of subsets of N, let I = {b N : (a I) a b = } and I + = {d N : (a I) d \ a I }. If J [N] , let J = JJ P(J). Say that J I is / unbounded in I if for each a I, there is a b J such that b \ a is innite. Denition 1.1. If I is an ideal of subsets of N, set cf(I) to be the conality of I; b(I) is the minimum cardinality of an unbounded family in I; (I) is the minimum cardinality of a subset J of I such that J is dense in I.
If N \ N = A x B, then IB = IA and x is the unique ultralter + + on N extending IA IB . The character of x in N \ N is equal to the maximum of cf(IA ) and cf(IB ).

modified:2007-08-12

revision:2007-08-12

Denition 1.2. Say that a tie-point x has (i) b-type; (ii) -type; respectively (iii) b-type, (, ) if N\N = A x B and (, ) equals: (i) (b(IA ), b(IB )) (ii) ((IA ), (IB )); and (iii) each of (b(IA ), b(IB )) and ((IA ), (IB )). We will adopt the convention to put the smaller of the pair (, ) in the rst coordinate. Again, it is interesting to note that if x is a tie-point of b-type (, ), then it is uniquely determined (in N \ N) by many subsets of N since x will be the unique point extending the family ((JA ) )+ ((JB ) )+ where JA and JB are unbounded subfamilies of IA and IB . Question 1.1. Can there be a tie-point in N \ N with -type (, ) with less than the character of the point?

916

TIE-POINTS AND FIXED-POINTS IN N

Question 1.2. Can N \ N have tie-points of -type (1 , 1 ) and (2 , 2 )? Proposition 1.2. If N \ N has symmetric tie-points of -type (, ) and (, ), but no tie-points of -type (, ), then N \ N has a 2-to-1 image which is not homeomorphic to N \ N. One could say that a tie-point x was radioactive in X (i.e. ) if X \ {x} can be similarly split into 3 (or more) relatively clopen sets accumulating to x. This is equivalent to X = A x B such that x is a tie-point in either A or B. Each point of character 1 in N \ N is a radioactive point (in particular is a tie-point). P-points of character 1 are symmetric tie-points of b-type (1 , 1 ), while points of character 1 which are not P-points will have b-type (, 1 ) and -type (1 , 1 ). If there is a tie-point of b-type (, ), then of course there are (, )-gaps. If there is a tie-point of -type (, ), then p . Proposition 1.3. If N \ N = A
x

B, then p (IA ).

mathp

modified:2007-08-12

Proof. If J IA has cardinality less than p, there is, by Solovays Lemma (and Bells Theorem) an innite set C N such that C and N\C each meet every innite set of the form J \( J ) where {J}J [J ]< . We may assume that C x, hence there are a IA and b IB / such that C a b. However no nite union from J covers a showing that J can not be dense in IA . Although it does not seem to be completely trivial, it can be shown that PFA implies there are no tie-points (the hardest case to eliminate is those of b-type (1 , 1 ))). Question 1.3. Does p > 1 imply there are no tie-points of b-type (1 , 1 )? Analogous to tie-points, we also dene a tie-set: say that K N\N is a tie-set if N \ N = A K B and K = A B, A = A \ K, and B = B \ K. Say that K is a symmetric tie-set if there is an involution F such that K = x(F ) and F [A] = B. Question 1.4. If F is an involution on N \ N such that K = x(F ) has empty interior, is K a (symmetric) tie-set? Question 1.5. Is there some natural restriction on which compact spaces can (or can not) be homeomorphic to the xed point set of some involution of N \ N? Again, we note a possible application to 2-to-1 maps.

916

revision:2007-08-12

A. DOW AND S. SHELAH

Proposition 1.4. Assume that F is an involution of N \ N with K = x(F ) = . Further assume that K has a symmetric tie-point x (i.e. K = A x B), then N \ N has a 2-to-1 continuous image which has a symmetric tie-point (and possibly N \ N does not have such a tie-point). Question 1.6. If F is an involution of N , is the quotient space N /F (in which each {x, F (x)} is collapsed to a single point) a homeomorphic copy of N \ N? Proposition 1.5 (CH). If F is an involution of N \ N, then the quotient space N /F is homeomorphic to N \ N. Proof. If x(F ) is empty, then N /F is a 2-to-1 image of N \ N, and so is a copy of N \ N. If x(F ) is not empty, then consider two copies, (N , F1 ) and (N , F2 ), of (N , F ). The quotient space of N /F1 N /F2 1 2 1 2 obtained by identifying the two homeomorphic sets x(F1 ) and x(F2 ) will be a 2-to-1-image of N , hence again a copy of N . Since N \x(F1 ) 1 and N \ x(F2 ) are disjoint and homeomorphic, it follows easily that 2 x(F ) must be a P-set in N . It is trivial to verify that a regular closed set of N with a P-set boundary will be (in a model of CH) a copy of N . Therefore the copy of N /F1 in this nal quotient space is a copy 1 of N .
modified:2007-08-12

2. a spectrum of tie-sets We adapt a method from [1] to produce a model in which there are tie-sets of specied b-types. We further arrange that these tie-sets will themselves have tie-points but unfortunately we are not able to make the tie-sets symmetric. In the next section we make some progress in involving involutions. Theorem 2.1. Assume GCH and that is a set of regular uncountable cardinals such that for each , T is a <-closed + -Souslin tree. There is a forcing extension in which there is a tie-set K (of b-type (c, c)) and for each , there is a tie-set K of b-type (+ , + ) such that K K is a single point which is a tie-point of K . Furthermore, for < c, if = or , then there is no tie-set of b-type / (, ). We will assume that our Souslin trees are well-pruned and are ever -ary branching. That is, if T is a + -Souslin tree, we assume that for each t T , t has exactly immediate successors denoted {t : } and that {s T : t < s} has cardinality + (and so has successors on every level). A poset is <-closed if every directed subset

main2

916

revision:2007-08-12

TIE-POINTS AND FIXED-POINTS IN N

of cardinality less than has a lower bound. A poset is <-distributive if the intersection of any family of fewer than dense open subsets is again dense. For a cardinal , let be the minimum cardinal such that ( )+ (i.e. the predecessor if is a successor). The main idea of the construction is nicely illustrated by the following. Proposition 2.2. Assume that N\N has no tie-sets of b-type (1 , 2 ) tie-sets for some 1 2 < c. Also assume that + < c is such that + is distinct from one of 1 , 2 and that T is a + -Souslin tree and {(at , xt , bt ) : t T } ([N] )3 satisfy that, for t < s T : (1) {at , xt , bt } is a partition of N, (2) xt j xt = for j < , (3) xs xt , at as , and bt bs , (4) for each , xt +1 at and xt +2 bt , + then if [T ] is a generic branch (i.e. () is an element of the -th level of T for each + ), then K = + x is a tie-set of () N \ N of b-type (+ , + ), and there is no tie-set of b-type (1 , 2 ). (5) Assume further that {(a , x , b ) : c} is a family of partitions of N such that {x : c} is a mod nite descending family of subsets of N such that for each Y N, there is a maximal antichain AY T and some c such that for each t AY , xt x is a proper subset of either Y or N\Y , then K = c x meets K in a single point z . (6) If we assume further that for each < < c, a a and b b , and for each t T , may be chosen so that xt meets each of (a \ a ) and (b \ b ), then z is a tie-point of K . Proof. To show that K is a tie-set it is sucient to show that K + + a + b . Since T is a -Souslin tree, no new subset of is added when forcing with T . Of course we use that is T is generic, so assume that Y N and that some t T forces that Y K is not empty. We must show that there is some t < s such that s forces that as Y and bs Y are both innite. However, we know that xt Y is innite for each since t T K xt . Therefore, by condition 4, for each , Y at and Y bt are both innite. Now let 1 , 2 be regular cardinals at least one of which is distinct from + . Recall that forcing with T preserves cardinals. Assume that in V [], K N and N = C K D with b(IC ) = (IC ) = 1 and b(ID ) = (ID ) = 2 . In V , let {c : 1 } be T -names for the increasing conal sequence in IC and let {d : 2 } be T -names for the increasing conal sequence in ID . Again using the fact that T adds

916

revision:2007-08-12

modified:2007-08-12

A. DOW AND S. SHELAH

modified:2007-08-12

no new subsets of N and the fact that every dense open subset of T will contain an entire level of T , we may choose ordinals { : 1 } and { : 2 } such that each t T , if t is on level it will force a value on c and if t is on level it will force a value on d . If 1 < + , then sup{ : 1 } < + , hence there are t T which force a value on each c . If + < 2 , then there is some < + , such that { 2 : } has cardinality 2 . Therefore there is some t T such that t forces a value on d for a conal set of 2 . Of course, if neither 1 nor 2 is equal to + , then we have a condition that decided conal families of each of IC and ID . This implies that N already has tie-sets of b-type (1 , 2 ). If 1 < 2 = + , then x t T deciding C = {c : 1 }, and let D = {d N : (s > t)s T d D}. It follows easily that D = C . But also, since forcing with T can not raise b(D) and can not lower (D), we again have that there are tie-sets of b-type in V . The case 1 = + < 2 is similar. Now assume we have the family {(a , x , b ) : c} as in (5) and (6) and set K = x , A = {K} {a : c}, and B = {K} {b : c}. It is routine to see that (5) ensures that the family {x x() : c and + } generates an ultralter when meets each maximal antichain AY (Y N). Condition (6) clearly ensures that A \ K and B \ K each meet (x x() ) for each c and + . Thus A K and B K witness that z is a tie-point of K . Let be a regular cardinal greater than + for all . We will need the following well-known Easton lemma (see [4, p234]). Lemma 2.3. Let be a regular cardinal and assume that P1 is a poset satisfying the -cc. Then any <-closed poset P2 remains <distributive after forcing with P1 . Furthermore any <-distributive poset remains <-distributive after forcing with a poset of cardinality less than . Proof. Recall that a poset P is <-distributive if forcing with it does not add, for any < , any new -sequences of ordinals. Since P2 is <-closed, forcing with P2 does not add any new antichains to P1 . Therefore it follows that forcing with P2 preserves that P1 has the -cc and that for every < , each -sequence of ordinals in the forcing extension by P2 P1 is really just a P1 -name. Since forcing with P1 P2 is the same as P2 P1 , this shows that in the extension by P1 , there are no new P2 -names of -sequences of ordinals.

distrib

916

revision:2007-08-12

TIE-POINTS AND FIXED-POINTS IN N

Now suppose that P2 is -distributive and that P1 has cardinality less than . Let D be a P1 -name of a dense open subset of P2 . For each p P1 , let Dp P2 be the set of all q such that some extension of p forces that q D. Since p forces that D is dense and that D Dp , it follows that Dp is dense (and open). Since P2 is -distributive, pP1 Dp is dense and is clearly going to be a subset of D. Repeating this argument for at most many P1 -names of dense open subsets of P2 completes the proof. We recall the denition of Easton supported product of posets (see [4, p233]). Denition 2.1. If is a set of cardinals and {P : } is a set of posets, then we will use P to denote the collection of partial functions p such that (1) dom(p) , (2) | dom(p) | < for all regular cardinals , (3) p() P for all dom(p). This collection is a poset when ordered by q < p if dom(q) dom(p) and q() p() for all dom(p). Lemma 2.4. For each cardinal , \+ T is <+ -closed and, if chain.condition is regular, T has cardinality at most 2< min( \ ). Lemma 2.5. If P is ccc and G P T is generic, then in V [G], chains for any and any family A [N] with |A| = : (1) if , then A is a member of V [G P ]; (2) if = + , , then there is an A A of cardinality + such that A is a member of V [G (P T )]; (3) if , then there is an A A of cardinality which is a / member of V [G P ]. Corollary 2.6. If P is ccc and G P T is generic, then for omit any < c such that either = or {+ : }, if there / is a tie-set of b-type (, ) in V [G], then there is such a tie-set in V [G P ]. Proof. Assume that N \ N = A K B in V [G] with = b(A) and = b(B). Let JA IA be an increasing mod nite chain, of order type , which is dense in IA . Similarly let JB IB be such a chain of order type . By Lemma 2.5, JA and JB are subsets of [N] V [GP ] = [N] . Choose, if possible 1 such that + = and 1 such that 1 + = . Also by Lemma 2.5, we can, by passing to a subcollection, 1 assume that JA V [G(P T1 )] (if there is no 1 , then let T1 denote
modified:2007-08-12

916

revision:2007-08-12

A. DOW AND S. SHELAH

the trivial order). Similarly, we may assume that JB V [G(P T1 )]. Fix a condition q G (P T ) which forces that (JA ) is a -dense subset of IA , that (JB ) is a -dense subset of IB , and that (IA ) = IB . Working in the model V [G P ] then, there is a family {a : } : 1 } of of T1 -names for the members of JA ; and a family {b T1 -names for the members of JB . Of course if = and T1 is the trivial order, then JA and JB are already in V [G P ] and we have our tie-set in V [G P ]. Otherwise, we assume that 1 < 1 . Set A to be the set of all a N such that there is some q(1 ) t T1 and such that t T1 a = a . Similarly let B be the set of all b N such that there is some q(1 ) s T1 and such that s T1 b = b . It follows from the construction that, in V [G], for any (a , b ) JA JB , there is an (a, b) A B such that a a and b b. Therefore the ideal generated by A B is certainly dense. It remains only to show that B (A) . Consider any (a, b) A B, and choose (q(1 ), q(1 )) (t, s) T1 T1 such that t T1 a JA and s T1 b JB . It follows that for any condition q q with q (P T ), q (1 ) = t, q (1 ) = s, we have that q
modified:2007-08-12

(P T )

a JA and b JB .

It is routine now to check that, in V [G P ], A and B generate ideals that witness that {(N\(ab)) : (a, b) AB} is a tie-set of b-type (, ). Let T be the rooted tree {} T and we will force an embedding of T into P(N) mod nite. In fact, we force a structure {(at , xt , bt ) : t T } satisfying the conditions (1)-(4) of Proposition 2.2. Denition 2.2. The poset Q0 is dened as the set of elements q = (nq , T q , f q ) where nq N, T q [T ]< , and f q : nq T q {0, 1, 2}. The idea is that xt will be qG {j nq : f q (j, t) = 0}, at will be q q qG {j n : f (j, t) = 1} and bt = N \ (at xt ). We set q < p if nq np , T q T p , f q f p and for t, s T p and i [np , nq ) (1) if t < s and f q (i, t) {1, 2}, then f q (i, s) = f q (i, t); (2) if t < s and f q (i, s) = 0, then f q (i, t) = 0; (3) if t s, then f q (i, t) + f q (i, s) > 0. (4) if j {1, 2} and{t , t ( +j)} T p and f q (i, t ( + j)) = 0, then f q (i, t ) = j. The next lemma is very routine but we record it for reference.

def.Q0

916

revision:2007-08-12

TIE-POINTS AND FIXED-POINTS IN N

Lemma 2.7. The poset Q0 is ccc and if G Q0 is generic, the family Q0 XT = {(at , xt , bt ) : t T } satises the conditions of Proposition 2.2. We will need some other combinatorial properties of the family XT . Denition 2.3. For any T [T ]< , we dene the following (Q0 names). (1) for i N, [i]T = {j N : (t T ) i xt i j xt }, (2) the collection n(T ) is the set of [i]T which are nite. ) n abbreviate n(T ) P(n). We abuse notation and let n(T Lemma 2.8. For each q Q0 and each T T q , n(T ) nq and for n i nq , [i]T is innite. Denition 2.4. A sequence SW = {(a , x , b ) : W } is a tower of T -splitters if for < W and t T : (1) {a , x , b } is a partition of N, (2) a a , b b , (3) xt x is innite. Denition 2.5. If SW is a tower of T -splitters and Y is a subset N, then the poset Q(SW , Y ) is dened as follows. Let EY be the (possibly empty) set of minimal elements of T such that there is some nite H W such that xt Y H x is nite. Let DY = EY = {t T : (s EY ) t s}. A condition q Q(SW , Y ) is a tuple (nq , aq , xq , bq , T q , H q ) where (1) nq N and {aq , xq , bq } is a partition of nq , (2) T q [T ]< and H q [W ]< , (3) (a \ a ), (b \ b ), and (x \ x ) are all contained in nq for < H q. We dene q < p to mean np nq , T p T q , H p H q , and (4) for t T p DY , xt (xq \ xp ) Y , (5) xq \ xp H p x , (6) aq \ ap is disjoint from bmax(H p ) , (7) bq \ bp is disjoint from amax(H p ) . Lemma 2.9. If W , SW is a tower of T -splitters, and if G is Q(SW , Y )-generic, then SW {(a , x , b )} is also a tower of T -splitters where a = {aq : q G}, x = {xq : q G}, and b = {bq : q G}. In addition, for each t DY , xt x Y (and xt x N \ Y for t EY ). Lemma 2.10. If W does not have conality 1 , then Q(SW , Y ) is -centered.

916

revision:2007-08-12

modified:2007-08-12

10

A. DOW AND S. SHELAH

As usual with (1 , 1 )-gaps, Q(SW , Y ) may not (in general) be ccc if W has a conal 1 sequence. Let 0 C be conal and assume that if C is conal in and / cf() = 1 , then C. poset Denition 2.6. Fix any well-ordering of H(). We dene a : } H(). We abuse nite support iteration sequence {P , Q notation and use Q0 rather than Q0 from denition 2.2. If C, / then let Q be the -least among the list of P -names of ccc posets in H() \ {Q : }. If C, then let Y be the -least P -name of a subset N which is in H() \ {Y : C }. Set Q to be the P name of Q(SC , Y ) adding the partition {a , x , b } and, where SC is the P -name of the T -splitting tower {(a , x , b ) : C }. We view the members of P as functions p with nite domain (or support) denoted dom(p). The main diculty to the proof of Theorem 2.1 is to prove that the iteration P is ccc. Of course, since it is a nite support iteration, this can be proven by induction at successor ordinals. main2ccc Lemma 2.11. For each C such that C has conality 1 , P+1 is ccc.
Proof. We proceed by induction. For each , dene p P if p P and there is an n N such that

modified:2007-08-12

(1) for each dom(p) C, with H = dom(p) C , there are subsets a , x , b of n and T [T ]< such that p P p() = (n, a , x , b , T , H )
Assume that P is dense in P and let p P+1 . To show that P+1 is / dense in P+1 we must nd some p p in P+1 . If C and p P is below p , then p {(, p()} is the desired element of P+1 . Now assume that C and assume that p P and that p forces that p() is the tuple (n0 , a, x, b, T , H). By an easy density argument, we may assume that H dom(p). Let n be the integer witnessing that p P . Let be the maximum element of dom(p) C and let p P p() = (n , a , x , b , T , H ) as per the denition of P+1 . Notice that since H H we have that

revision:2007-08-12

(n , a , x, b , T T , H {}) p()

where a = a ([n0 , n ) \ b ) and b = b ([n0 , n ) b ). Dening p P+1 by p = p and p () = (n , a , x, b , T T , H {})

916

TIE-POINTS AND FIXED-POINTS IN N

11

completes the proof that P+1 is dense in P+1 , and by induction, that this holds for = . Now assume that {p : 1 } P+1 . By passing to a subcollection, we may assume that (1) the collection {T p () : 1 } forms a -system with root T ; (2) the collection {dom(p ) : 1 } also forms a -system with root R; (3) there is a tuple (n , a , x , b ) so that for all 1 , ap () = a , xp () = x , and bp () = b . Since C has a conal sequence of order type 1 , there is a such that R and, we may assume, (dom(p )\) min(dom(p )\) for < < 1 . Since P is ccc, there is a pair < < 1 such that p is compatible with p . Dene q P+1 by (1) q is any element of P which is below each of p and p , (2) if dom(p ), then q() = p (), (3) if dom(p ) \ C, then q() = p (), (4) if dom(p ) C, then

q() = (n , ap () , xp () , bp () , T p () , H p () H p () ). The main non-trivial fact about q is that it is in P+1 which depends on the fact that, by induction on C , q forces that
modified:2007-08-12

(a \ a ) (b \ b ) (x \ x ) n for C . It now follows trivially that q is below each of p and p . Proof of Theorem 2.1. This completes the construction of the ccc poset P (P as above). Let G (P T ) be generic. It follows that V [G P ] is a model of Martins Axiom and c = . Furthermore by applying Lemma 2.4 with = and Lemma 2.3, we have that P2 = T is 1 -distributive in the model V [G P ]. Therefore all subsets of N in the model V [G] are also in the model V [G P ]. Fix any and let denote the generic branch in T given by G. Let G denote the generic lter on P {T : = } and work in the model V [G ]. It follows easily by Lemma 2.4 and Lemma 2.3, that T is a + -Souslin tree in this model. Therefore by Proposition 2.2, K = <+ x () is a tie-set of b-type (+ , + ) in V [G]. By the denition of the iteration in P , it follows that condition (4) of Lemma 2.2 is also satised, hence the tie-set K = C x meets K in a single point z . A simple genericity argument conrms that conditions (5) and (6) of Proposition 2.2 also holds, hence z is a tie-point of K .

916

revision:2007-08-12

12

A. DOW AND S. SHELAH

It follows from Corollary 2.6 that there are no unwanted tie-sets in N \ N in V [G], at least if there are none in V [G P ]. Since p = c in V [G P ], it follows from Proposition 1.3 that indeed there are no such tie-sets in V [G P ]. Unfortunately the next result shows that the construction does not provide us with our desired variety of tie-points (even with variations in the denition of the iteration). We do not know if b-type can be improved to -type (or simply exclude tie-points altogether). Proposition 2.12. In the model constructed in Theorem 2.1, there are no tie-points with b-type (1 , 2 ) for any 1 2 < c, Proof. Assume that N\N = A x B and that (IA ) = 1 and (IB ) = 2 . It follows from Corollary 2.6 that we can assume that 1 = 2 = + for some . Also, following the proof of Corollary 2.6, there are P T -names JA = { : + } and P T+ -names JB = { : a b + } such that the valuation of these names by G result in increasing (mod nite) chains in IA and IB respectively whose downward closures are dense. Passing to V [GP ], since T has the -cc, there is a Boolean subalgebra B [P(N)]< such that each a and is a name of a b member of B. Furthermore, there is an innite C N such that C x / and each of b C and b \ C are innite for all b B. Since C x, there / is a Y N (in V [G]) such that C Y IA and C \ Y IB . Now choose t0 T which forces this about C and Y . Back in V [G P ], set A = {b B : (t1 t0 ) t1
T

modified:2007-08-12

b JA JB } .

Since V [G P ] satises p = and A is forced by t0 to be dense in [N] , there must be a nite subset A of A which covers C. It also follows easily then that there must be some a, b A and t1 , t2 each below t0 such that t1 T+ a JA , t2 T+ b JB , and a b is innite. The nal contradiction is that we will now have that t0 fails to force that C a Y and C b (N \ Y ). 3. T -involutions In this section we strengthen the result in Theorem 2.1 by making each K K a symmetric tie-point in K (at the expense of weakening Martins Axiom in V [G P ]). This is progress in producing involutions with some control over the xed point set but we are still not able to make K the xed point set of an involution. A poset is said to be linked if there is a countable collection of linked (elements are pairwise compatible) which union to the poset. The statement MA( linked)

916

revision:2007-08-12

TIE-POINTS AND FIXED-POINTS IN N

13

is, of course, the assertion that Martins Axiom holds when restricted to -linked posets. Our approach is to replace T -splitting towers by the following notion. If f is a (partial) involution on N, let min(f ) = {n N : n < f (n)} and max(f ) = {n N : f (n) < n} (hence dom(f ) is partitioned into min(f ) x(f ) max(f )). Denition 3.1. A sequence T = {(A , f ) : W } is a tower of T -involutions if W is a set of ordinals and for < W and t T (1) A A ; 2 (2) f = f and f (N \ x(f )) f ; (3) f [xt ] = xt and x(f ) xt is innite; (4) f ([n, m)) = [n, m) for n < m both in A . Say that T, a tower of T -involutions, is full if K = KT = {x(f ) : W } is a tie-set with N\N = A K B where A = K {min(f ) : W } and B = K {max(f ) : W }. If T is a tower of T -involutions, then there is a natural involution FT on W (N \ x(f )) , but this FT need not extend to an involution on the closure of the union - even if the tower is full. In this section we prove the following theorem. Theorem 3.1. Assume GCH and that is a set of regular uncountable main3 cardinals such that for each , T is a <-closed + -Souslin tree. Let T denote the tree sum of {T : }. There is forcing extension in which there is T, a full tower of T -involutions, such that the associated tie-set K has b-type (c, c) and such that for each , there is a tie-set K of b-type (+ , + ) such that FT does induce an involution on K with a singleton xed point set {z } = K K . Furthermore, for < c, if = or , then there is no tie-set of b-type / (, ). Question 3.1. Can the tower T in Theorem 3.1 be constructed so that FT extends to an involution of N \ N with x(F ) = KT ?

revision:2007-08-12

modified:2007-08-12

We introduce T -tower extending forcing. Denition 3.2. If T = {(A , f ) : W } is a tower of T -involutions tower and Y is a subset of N, we dene the poset Q = Q(T, Y ) as follows. Let EY be the (possibly empty) set of minimal elements of T such that there is some nite H W such that xt Y H x(f ) is nite. Let DY = EY = {t T : (s EY ) t s}. A tuple q Q if q = (aq , f q , T q , H q ) where:

916

14

A. DOW AND S. SHELAH

(1) H q [W ]< , T q [T ]< , and nq = max(aq ) Aq where q = max(H q ), (2) f q is an involution on nq , (3) (Aq \ nq ) A for each H q , (4) n(T q ) nq , (5) f (N \ (x(f ) nq )) fq for H q , (6) fq [xt \ nq ] = xt \ nq for t T q , We dene p < q if np nq , and for t T p and i [np , nq ): (7) ap = aq np , T p T q , and H p H q , (8) aq \ ap Ap , (9) fp (i) = i implies f q (i) = fp (i), (10) f q ([n, m)) = [n, m) for n < m both in aq \ ap , (11) f q (xt [np , nq )) = xt [np , nq ), (12) if t D p and i xt x(f q ), then i Y It should be clear that the involution f introduced by Q(T, Y ) satises that for each t DY , x(f ) xt Y , and, with the help of the following density argument, that T {(, A, f )} is again a tower of T -involutions where A is the innite set introduced by the rst coordinates of the conditions in the generic lter. density
modified:2007-08-12

Lemma 3.2. If W , Y N, and T = {(A , f ) : W } is a tower of T -involutions and p Q(T, Y ), then for any T [T ]< , W , and any m N, there is a q < p such that nq m, H q , T q T , and q p p x(f ) (xt \ n ) is not empty for each t T . Proof. Let denote the maximum p and and let denote the minimum. Choose any nq Aq \ m large enough so that (1) fp [xt \ nq ] = xt \ nq for t T , q (2) f (N \ (n x(f ))) f , (3) A \ A is contained in nq , (4) nq [i]T p x(fp ) is non-empty for each i N such that [i]T p is in the nite set {[i]T p : i N} \ n(T p ), (5) if i xt nq \np for some t DY T p , then Y meets [i]T p nq \np in at least two points. Naturally we also set H q = H p {} and T q = T p T . The choice q of n is large enough to satisfy (3), (4), (5) and (6) of Denition 3.2. We will set aq = ap {nq } ensuring (1) of Denition 3.2. Therefore for any f q f p which is an involution on nq , we will have that q = (aq , f q , T q , H q ) is in the poset. We have to choose f q more carefully to ensure that q p. Let S = [np , nq ) x(fp ), and S = [np , nq ) \ S. We choose f an involution on S and set f q = f p (fp S ) f . We leave

916

revision:2007-08-12

TIE-POINTS AND FIXED-POINTS IN N

15

it to the reader to check that it suces to ensure that f sends [i]T p S p to itself for each t T and that x(f ) xt Y for each t T p DY . Since the members of {[i]T p S : i N} are pairwise disjoint we can dene f on each separately. For each [i]T p S which has even cardinality, choose two points yi , zi from it so that if there is a p DY T p such that [i]T p xt , then {yi , zi } Y . Let f be any involution on [i]T p S so that yi , zi are the only xed points. If [i]T p S has odd cardinality then choose a point p yi from it so that if [i]T is contained in xt for some t Dy T p , then yi Y [i]T p S. Set f (yi ) = yi and choose f to be any xed-point Tp free involution on [i] S \ {yi }. Let P now be the nite support iteration dened as in Denition 2.6 except for two important changes. For C, we replace T -splitting towers by the obvious inductive denition of towers of T -involutions when we replace the posets Q(SC , Y ) by Q(TC , Y ). For C we / require that P Q is -linked. Special (parity) properties of the family {xt : t T } are needed to ensure that P Q(SC , Y ) is ccc even for cases when cf() is not 1 . The proof of Theorem 3.1 is virtually the same as the proof of Theorem 2.1 (so we skip) once we have established that the iteration is ccc.
modified:2007-08-12

Lemma 3.3. For each C, P+1 is ccc.

Proof. We again dene P to be those p P for which there is an n N such that for each dom(p) C, there are n a n+1, f nn , T [T ]< , and H = dom(p) C such that p P p() = (a , f , T , H ). However, in this proof we must also make some special assumptions in coordinates other than those in C. For each \ C, we x a collection {Q(, n) : n } of P -names so that 1 P Q = Q(, n) and (n) Q(, n) is linked. n The nal restriction on p P is that for each \ C, there is a k such that p P p() Q(, k ). Just as in Lemma 2.11, Lemma 3.2 can be used to show by induction that P is a dense subset of P . This time though, we also demand that dom(f p(0) ) = n T p(0) is such that T T p(0) for all dom(p) C and some extra argument is needed because of needing to decide values in the name Y as in the proof of Lemma 3.2. Let p P+1 and assume that P is dense in P . By density, we may assume that p

main3ccc

916

revision:2007-08-12

16

A. DOW AND S. SHELAH

P , H p() dom(p), T p() T p(0) , and that p has decided the members of the set DY T p() . We can assume further that for p() each t DY T , p has forced a value yt Y xt \ {xs : s T p and s t} such that yt > np() . We are using that T is not nitely branching to deduce that if t DY , then p P Y xt \ {xs : s T p and s t} is non-empty (which follows since Y must meet xs for each immediate successor s of t). Choose any m larger than yt for each t T p() . Without loss of generality, we may assume that the integer n witnessing that p P is at least as large as m and that n H p() A . Construct f just as in Lemma 3.2, except that this time there is no requirement to actually have xed points so one member of Y in each appropriate [i]T p() is all that is required. Let = max(dom(p) ). No new forcing decisions are required of p in order to construct a suitable f , hence this shows that p {(, q)} (where q is constructed below p() as in Corollary 3.2 in which H p() {} is add to H q ) is the desired extension of p which is a member of P+1 . Now to show that P+1 is ccc, let {p : 1 } P+1 . Clearly we may assume that the family {p (0) : 1 } are pairwise compatible and that there is a single integer n such that, for each 1 , dom(p (0)) = n T for some T [T ]< . Also, we may assume that there is some (a, h) such that, for each ,
modified:2007-08-12

p() = (a, h, T , H )

revision:2007-08-12

where H = dom(p ) C . The family {dom(p ) : 1 } may be assumed to form a system with root R. For each R, we may assume that, if C, / there is a single k such that, for all , p P p () Q(, k ), and if C, then there is a single (a , h ) such that p P p () = (a , h , T , H ). For convenience, for each C / 2 let r be a P -name of a function from Q such that, for each k , 1 r (k, q, q ) q, q (q, q Q(, k)).

Fix any < < 1 and let H = H H . Recall that p (0) and p (0) are compatible. Recursively dene a P -name q() for

916

TIE-POINTS AND FIXED-POINTS IN N

17

modified:2007-08-12

Now we check that q P by induction on + 1. The rst thing to note is that not only is this true for = 1, but also that q(0) Q0 n(T T ) n. Since p and p are each in P+1 , this show that condition (4) of Denition 3.2 will hold in all coordinates in C. We also prove, by induction on , that q forces that for < both in H and t T T , f [xt \ n] = xt \ n, f (N \ (x(f ) n)) f and A \ n A . Given H and the assumption that q P , and = q() = max(H ), condition (3), (5), and (6) of Denition 3.2 hold by the inductive hypothesis above. It follows then that q P q() Q . By the denition of the ordering on Q , given that H = H q() and T T = T q() , it follows that the inductive hypothesis then holds for + 1. It is trivial for dom(q) \ C, that q P implies that q P q() Q . This completes the proof that q P+1 , and it is trivial that q is below each of p and p . Remark 1. If we add a trivial tree T1 to the collection {T : } (i.e. T1 has only a root), then the root of T has a single extension which is a maximal node t, and with no change to the proof of Theorem 3.1, one obtains that F induces an automorphism on x with a single xed t point. Therefore, it is consistent (and likely as constructed) that N\N will have symmetric tie-points of type (c, c) in the model V [G P ] and V [G]. Remark 2. In the proof of Theorem 2.1, it is easy to arrange that each K ( ) is also KT for a (T -generic) full tower, T , of Ninvolutions. However the generic sets added by the forcing P will prevent this tower of involutions from extending to a full involution. 4. Questions In this section we list all the questions with their original numbering.

dom(p ) dom(p ) so that q P (n, T T , f p (0) f p (0) ) r (k , p (), p ()) q() = p () p () (a , h , T T , H )

=0 R\C dom(p ) \ (R C) . dom(p ) \ (R C) C.

916

revision:2007-08-12

18

A. DOW AND S. SHELAH

Question 1.1. Can there be a tie-point in N \ N with -type (, ) with less than the character of the point? Question 1.2. Can N \ N have tie-points of -type (1 , 1 ) and (2 , 2 )? Question 1.3. Does p > 1 imply there are no tie-points of b-type (1 , 1 )? Question 1.4. If F is an involution on N \ N such that K = x(F ) has empty interior, is K a (symmetric) tie-set? Question 1.5. Is there some natural restriction on which compact spaces can (or can not) be homeomorphic to the xed point set of some involution of N \ N? Question 1.6. If F is an involution of N , is the quotient space N /F (in which each {x, F (x)} is collapsed to a single point) a homeomorphic copy of N \ N?
modified:2007-08-12

Question 3.1. Can the tower T in Theorem 3.1 be constructed so that FT extends to an involution of N \ N with x(F ) = KT ? References
[1] Jrg Brendle and Saharon Shelah, Ultralters on their ideals and their caro dinal characteristics, Trans. Amer. Math. Soc. 351 (1999), no. 7, 26432674. MR 1686797 (2000m:03111) [2] Alan Dow, Two to one images and PFA, Israel J. Math. 156 (2006), 221241. MR 2282377 [3] Alan Dow and Geta Techanie, Two-to-one continuous images of N , Fund. Math. 186 (2005), no. 2, 177192. MR 2162384 (2006f:54003) [4] Thomas Jech, Set theory, Springer Monographs in Mathematics, SpringerVerlag, Berlin, 2003, The third millennium edition, revised and expanded. MR 1940513 (2004g:03071) [5] Ronnie Levy, The weight of certain images of , Topology Appl. 153 (2006), no. 13, 22722277. MR MR2238730 (2007e:54034) [6] S. Shelah and J. Steprns. Non-trivial homeomorphisms of N \ N without the a Continuum Hypothesis. Fund. Math., 132:135141, 1989. [7] S. Shelah and J. Steprns. Somewhere trivial autohomeomorphisms. J. London a Math. Soc. (2), 49:569580, 1994.

916

revision:2007-08-12

TIE-POINTS AND FIXED-POINTS IN N

19

[8] Saharon Shelah and Juris Steprns, Martins axiom is consistent with the exa istence of nowhere trivial automorphisms, Proc. Amer. Math. Soc. 130 (2002), no. 7, 20972106 (electronic). MR 1896046 (2003k:03063) [9] B. Velikovi. Denable automorphisms of P()/f in. Proc. Amer. Math. Soc., c c 96:130135, 1986. [10] Boban Velikovi. OCA and automorphisms of P()/n. Topology Appl., c c 49(1):113, 1993.

Department of Mathematics, Rutgers University, Hill Center, Piscataway, New Jersey, U.S.A. 08854-8019 Current address: Institute of Mathematics, Hebrew University, Givat Ram, Jerusalem 91904, Israel E-mail address: shelah@math.rutgers.edu

916

revision:2007-08-12

modified:2007-08-12