You are on page 1of 9

Open Problems in Topology J. van Mill and G.M. Reed (Editors) c Elsevier Science Publishers B.V.

(North-Holland), 1990

Chapter 16 Large Homogeneous Compact Spaces Kenneth Kunen1


Department of Mathematics University of Wisconsin Madison, WI 53706, U.S.A. kunen@cs.wisc.edu

Contents 1. The Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263 2. Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270

1 This

research was supported by NSF Grant DMS-8801139.

1. The Problem Throughout this paper, space always means Hausdor space. Most of the spaces considered are compact. We use 2 to denote the 2-point space, {0, 1}, with the (necessarily) discrete topology, and [0, 1] to denote the unit interval with the usual topology. The cellularity of a space is the supremum of all sizes of families of disjoint open sets in the space; thus, a space is ccc i its cellularity is 0 or less. The weight of a space is the least cardinality of a basis, and the character of a point in a space is the least cardinality of a local base at that point. A space, X, is called homogeneous i for any x, y X, there is a homeomorphism, h, of X onto X, such that h(x) = y. For example, 2 (with the product topology) is homogeneous for any cardinal . Thus, homogeneous compact spaces can be made arbitrarily large in the sense of the cardinal functions cardinality, weight, and character; however the spaces 2 are small in the sense of cellularity, since they all are ccc. There are also non-ccc compact homogeneous spaces; for example, let X be the space 2 for an ordinal , where now X has the order topology, using lexicographical order. By Maurice [1964], if is countable and indecomposable ( < ( + = )), then X is homogeneous. If = , then X is just the Cantor set, but if > (for example, is the ordinal ), then X has cellularity 20 . However, since really must be countable here, 20 is the largest cellularity obtainable by this method. Thus, a natural question, posed rst by van Douwen, is: Problem. Is there a compact homogeneous space with cellularity greater 391. ? than 20 ? Perhaps this question depends on the axioms of set theory, although as far as I know, the problem is open under any axioms. In the following, I survey some partial results, with the hope that it might be helpful in the solution. For a much more detailed survey, see Arkhangel ski [1987]. If the answer is yes, you need a non-trivial technique for producing homogeneous spaces; unfortunately, I dont know of any that seem helpful here. There are a number of results on products. Trivially, the product of homogeneous spaces is homogeneous. It is true (Keller [1931]), and non-trivial, that [0, 1] is homogeneous for any innite , but these all have the ccc. Likewise, Motorov showed that X is homogeneous whenever is innite and X is compact, rst countable, and 0-dimensional, but the cellularity of such a product cannot exceed 20 . It is easy to produce homogeneous Boolean algebras, but their Stone spaces are not in general homogeneous (see, e.g., below). Ordered spaces wont work; in fact, every homogeneous compact LOTS is rst countable (and hence has cardinality (and hence cellularity) no more than 20 ). For the proof, observe, by taking a nested sequence of intervals, 263

264

Kunen / Large Homogeneous Compact Spaces

[ch. 16

that every compact LOTS contains either a P -point or a point of countable character. Thus, if X is a homogeneous compact LOTS, then either every point is a P -point (so X is nite), or every point has character 0 . If the answer is no, you need a non-trivial technique for proving compact spaces nonhomogeneous. One such technique, due independently to M. E. Rudin and Z. Frol establishes that N is not homogeneous. Here, N dek, notes with the discrete topology, N is its Cech compactication, and N is the remainder, N\N. We identify each point p N with a non-principal ultralter. If xn (n ) are points in any compact space, X, and p N , we dene the p-limit, limp xn : n , to be the (unique) y X such that for each neighborhood, U , of y, {n : xn U } p (viewing p as an ultralter). Then, in N , they showed that for any such p, p is not a p-limit of any discrete sequence of points. Thus, taking y to be a p-limit of a discrete sequence, no homeomorphism of N can move p to y. The property which distinguishes p from y in the above proof (being a plimit of a discrete sequence) is a little complicated. Under the Continuum Hypothesis or Martins Axiom, a simpler proof would be to use W. Rudins theorem that there is a P -point in N ; however, Shelah showed that one cannot prove outright in ZFC that there is such a P -point. However, there is always a weak P -point. In general, a point, x X is called a weak P -point i x is not a limit of any countable subset of X. I proved in Kunen [1978] that there is a weak P -point in N , a result which has since been greatly improved by van Mill and others; see van Mill [1984]. Since every innite compact space must also contain points which are not weak P -points, this establishes the nonhomogeneity of N via a more quotable property. The Rudin-Frol proof is still of great importance, since their method k applies to spaces which do not contain weak P -points. For example, one can use the study of p-limits to prove that no innite compact F -space is homogeneous (or even stronger resultssee Section 2). Here, X is called a compact F -space i X is compact and in X, any two disjoint open F sets have disjoint closures. For example, N is a compact F -space. Not every compact F -space has weak P -points; for example, the absolute of [0, 1] (equivalently, the Stone space of the regular open algebra of [0, 1]) is separable, and thus has no weak P -points. The Stone space of any complete Boolean algebra is a compact F -space; this provides a large class of examples of homogeneous Boolean algebras whose Stone spaces are not homogeneous. Conversely, van Douwen [1981] has shown that there is a non-homogeneous Boolean algebra whose Stone space is homogeneous. Returning to the problem of homogeneous compact spaces with large cellularity, it might be hoped that one might do this by taking products, in analogy to the results mentioned earlier for rst countable spaces. If so, the factors must be chosen with some care as we show in Section 2, for example, no prod-

2]

Products

265

uct of innite compact F -spaces is homogeneous. So, it now seems natural to consider the following classes of compact spaces, X, graded by successively weaker homogeneity properties. Class 1. X is homogeneous. Class 2. For some compact Y , X Y is homogeneous. Class 3. X is a retract of a compact homogeneous space. Class 4. X is a continuous image of a compact homogeneous space. Perhaps Class 4 contains all compact spaces. Of course, if it contains any compact space of cellularity greater than 20 , then the answer to the Problem is yes. Class 3 does not contain all compact spaces. If X is the closure in the plane of the graph of sin(1/x), x (0, 1], then Motorov has shown, by a connectedness argument, that X is not a retract of any compact homogeneous space; see Arkhangel ski [1987] for a more general result along this line. Note that this X is a continuous image of the Cantor set, and hence in Class 4. It is not clear whether Classes 2 and 3 are distinct, or whether they contain all compact 0-dimensional spaces. By Motorovs previously mentioned result, every compact, 0-dimensional, rst countable space is in Class 2. It is also not clear whether Class 2 contains any innite compact F -space; if so, then the proof in section 2 puts some restrictions about what the Y can be. Specically, Y cannot be any compact LOTS or any compact metric space or any compact F -space, or any product of such spaces. Finally, a simple sequence is a trivial example to show that Classes 1 and 2 are dierent, even for metric spaces. 2. Products The results of this section show that any product of an innite compact F space with any collection of other compact F -spaces or simple compact spaces is not homogeneous. I am not sure what the best denition of simple is to get the strongest result (perhaps simple = any), but it certainly includes all sequentially compact spaces. A space, X, is called sequentially compact i in X, every -sequence has a convergent subsequence. So, 2 is sequentially compact, as is any compact metric space. Every compact LOTS is sequentially compact (just choose the subsequence to be increasing or decreasing). Many compact spaces fail to be sequentially compact; for example, N, or N , or 2 for any 20 ; our Theorem below will, however, apply to 2 , since it allows arbitrary products of sequentially compact spaces. Martins Axiom implies that every compact space of weight less than 20 is sequentially compact. This is not provable in ZFC, although our Theorem will apply to such spaces anyway by the following extension, which we now discuss. Call a space, X, sequentially small i whenever A is an innite subset

266

Kunen / Large Homogeneous Compact Spaces

[ch. 16

of X, there is an innite B A whose closure does not contain a copy of N. Obviously, every compact sequentially compact space and every space of weight less than 20 is sequentially small. In fact, if all points of X have character less than 20 , then X is sequentially small, since by a theorem of Pospsil (see van Mill [1984]), N must contain a point of character 20 . No innite compact F -space can be sequentially small, since in such a space the closure of every innite set contains a copy of N. Then, our result is: 1. Theorem. Suppose X = < X , where each X is either an innite compact F -space or contains a weak P -point or has a non-empty sequentially small open subset. Suppose further that at least one X is an innite compact F -space. Then X is not homogeneous. We begin with some preliminaries on products. If we have a product, X = < X , we shall use subscripts for the coordinates, and superscripts for indices of sequences in X. Thus, we might consider a sequence, xn : n from X; each xn would be a -sequence xn : , where xn X . If S , we let S be the natural projection from X onto S X . Now, let us look more closely at sequences. In any compact space, call the sequence dn : n nicely separated i the dn are all distinct and there are open neighborhoods, U n (n ) of the dn such that for all A , cl(
nA

U n ) cl(
nA /

U n) = .

We say that the U n nicely separate the dn . Nicely separated is the opposite of being convergent; it implies that the closure of {dn : n } is homeomorphic to N. If X is an F -space, every discrete -sequence is nicely separatedjust take the U n to be disjoint open F sets. By the next lemma, in a product space, nicely separated depends only on countable subproducts. 2. Lemma. Suppose X = < X , where each X is compact, and suppose each dn X (n ). Then in the following, (a) (b) and (b) (c): (a) For some , dn : n is nicely separated in X . (b) For some countable S , S (dn ) : n is nicely separated in S X . (c) dn : n is nicely separated in X. Proof. It is easy to see that (a) (b) and (b) (c); just pull back the nicely separating neighborhoods. To show (c) (b), let the U n nicely separate the dn . We may, by shrinking them if necessary, assume that each U n is an open F set; say U n = k F n,k , where each F n,k is closed. But then, by a standard compactness argument, each U n is a cylinder over a countable set of co-ordinates. That is, each F n,k is covered by nitely many basic subsets of

2]

Products

267

U n ; taking the union of the supports of these basic sets for all n, k produces 1 a countable S such that each U n = S S (U n ). Then the S (dn ) are nicely separated by the S (U n ). We remark that in Lemma 2, it is easy to produce examples where (b) does not imply (a), even when = 2. Next, we look more closely at weak P -points in N . As before, we identify N with the set of ultralters on and N with the set of non-principal ultralters. If p N and : , we dene (p) N by: A (p) 1 (A) p. If p, q N , we dene p q i there is a function, : such that p = (q). It is easy to verify that is transitive and reexive. This partial order is due to M. E. Rudin and H. J. Keisler. We call p and q Rudin-Keisler incomparable i p q and q p. Then we quote the following lemma from Kunen [1978]: 3. Lemma. There are weak P -points, p and q in N such that p and q are Rudin-Keisler incomparable. The next lemma uses the method of M. E. Rudin and Z. Frol to show that k in a compact F -space, a p-limit of a discrete sequence cannot be a q-limit of any sequence, except in the trivial case of a constant sequence. 4. Lemma. Suppose p, q N are weak P -points and are Rudin-Keisler incomparable. Let X be any compact F -space. In X, let dm : m be a discrete sequence of distinct points, and en : n any sequence of points (possibly not distinct). Suppose that x = limp dm : m = limq en : n . Then {n : en = x} q . Proof. Let K be the closure of {dm : m }, and K = K\{dm : m }. Choose open sets, U m (m ) so that each dm U m , and the U m are disjoint from K and from each other. Let A = {n : en K }, B = {n : en m m U }, and C = \(A B). Then one of A, B, C is in q, resulting in three cases; the second and third will lead to contradictions. Case 1: A q: Dene a map f on N by f (n) = dn . Then, by the properties of the Cech compactication, f extends to a map, which we also call f , from N onto K. Since X is an F -space, f is a homeomorphism, let g be its inverse. If n A, then g(en ) N , and g(x) = p = limq g(en ) : n . Since p is a weak P -point, we must have that {n : en = x} = {n : g(en ) = p} q . Case 2: B q: For n B, let (n) be that m such that en U m . Then p = (q), contradicting that p and q were Rudin-Keisler incomparable. Case 3: C q: Observe that for n C, en K. By induction on n , / choose open F sets, V n and W n , such that for each n, (i) dn V n cl(V n ) U n . (ii) If n C, then en W n . (iii) cl(W n ) is disjoint from K and from V i for all i n (If n C we can / take W n = ).

268

Kunen / Large Homogeneous Compact Spaces

[ch. 16

(iv) V n is disjoint from W i for all i < n; this is possible by (iii). But then n V n and n W n are disjoint open F sets and x is in the closure of both of them (by (ii)), contradicting that X is an F -space. Of course, it follows that no innite compact F -space can be homogeneous, since a p-limit of a discrete sequence cannot be moved to the q-limit of any discrete sequence by a homeomorphism. The two distinct types of points we produce, p-limit of discrete sequence and q-limit of discrete sequence, are not very quotable, but the types produced in the proof of our main Theorem will be even less quotable. Lemma 4 will be false in a product of two compact F -spaces, since in such a product a p-limit of a discrete sequence along the x-axis can equal a q-limit of a discrete sequence along the y-axis. Thus, a more complex argument is needed to refute the homogeneity of such a product. First, we need to quote one more lemma, due to Malykhin [1979]. 5. Lemma. If N is embeddable in least one Xi .
i

Xi , then N is embeddable in at

Proof. (Theorem 1) The hypotheses on the X are not mutually exclusive, but partition arbitrarily into 3 subsets, R, S and T , such that R = , and such that each X is an innite compact F -space for R, contains a weak P -point for S, and contains a non-empty sequentially small open set for T. Choose dn X for n as follows: For each R let dn : n be a discrete sequence in X . For each S let the dn be all the same weak P -point in X . For each T let U be a non-empty open subset of X whose closure is sequentially small, and let the dn be all the same element of U . Let p and q be as in Lemma 3. Let x = limp dm : m and y = limq dm : m . We assume that h is a homeomorphism of X with h(y) = x, and derive a contradiction. Let en = h(dn ). The dn are nicely separated in X since R = ; thus the n e are also nicely separated in X. Applying Lemma 1, x a countable J such that J (en ) : n is nicely separated in J X . Observe that for each , x = limq em : m = limp dm : m . We consider the three kinds of spaces X separately. First, for each J R, we may apply Lemma 4 to X and choose an A q so that em = x for all m A . We may do likewise for each J S, using the fact that x is a weak P -point. Since J (R S) is countable, we may now choose an innite B which is almost contained in each of these A ; so, for these , em = x for all but at most nitely many m B. So, for these , {em : m B} is nite. We are not claiming that B q. Next, for those J T , apply the denition of sequentially small times and diagonalize to get an innite D B so that for each such , cl({em : m D}) does not embed N. (At each application, one rst tries

2]

Products

269

to nd a subsequence where the em are distinct; if this is impossible, then {em : m D} will be nite.) For each J, let P = cl({em : m D}). Then each P does not embed N, so neither does J P by Lemma 5. But this is a contradiction; the product contains the closure of {J (en ) : n D}, which is homeomorphic to N, since the points are nicely separated. If one assumes the Continuum Hypothesis or Martins Axiom, some of the minor steps along the way to the proof can be strengthened. Now, one can take p and q to be selective; i.e., minimal in the Rudin-Keisler order. In the proof of the Theorem, we can then indeed take B q. Also, in Lemma 4, we do not have to assume that the sequence dm : m is discrete; it is enough to assume the dm are distinct, since then there will be an E p such that dm : m E is discrete anyway. However, it is not clear whether the Theorem itself can be strengthened in any essential way. The Theorem overlaps somewhat with results of van Douwen [1978], who established nonhomogeneity by a dierent methodconsidering cardinal functions rather than limit points. He showed, for example, that if X is compact and |X| > 2(X) , then no power of X is homogeneous. So, this gives a dierent proof that no power of N is homogeneous. Another method, emphasizing limit points again, also may be used to establish nonhomogeneity of products in some cases. Call a strict F1 any union of the form {U : < 1 }, where each U is open and contains the closure of {U : < }. Call x an L-point of X i x is in the boundary of a strict F1 . Observe that if X is a product of spaces, X , then x is an L-point of X i some x is an L-point of X . Thus, if each X contains a non-L-point and some X contains an L-point, then the product is not homogeneous. So, for example, the ordinal 1 + 1 cross any product of ccc spaces and spaces containing a point of countable tightness is nonhomogeneous. Yet another nonhomogeneity result is due to Dow and van Mill [1980], who showed that no compact space can be covered by nowhere dense ccc P -sets. Thus, if X is compact and contains a non-isolated P -point, then X cross any compact ccc space is not homogeneous. This applies to X = 1 + 1, although here, the L-point argument gives a stronger result. More interestingly, it applies when X is the LOTS, 21 , ordered lexicographically; here, every point is an L-point, so that the L-point argument says nothing. It seems that the state of the art on homogeneity of products can be summarized by saying that there are a lot of special results, without any unifying theme emerging yet.

270

Kunen / Large Homogeneous Compact Spaces

[ch. 16

References
Arkhangel ski , A. V. [1987] Topological homogeneity. Topological groups and their continuous images. Russian Math. Surveys, 42, 83131. Russian original: Uspehi Mat. Nauk, 42 (1987), 69-105. van Douwen, E. K. [1978] Nonhomogeneity of products of preimages and -weight. Proc. Amer. Math. Soc., 69, 183192. [1981] A compact space with a measure that knows which sets are homeomorphic. Adv. Math., 52, 133. Dow, A. and J. van Mill. [1980] On nowhere dense ccc P -sets. Proc. Amer. Math. Soc., 80, 697700. Keller, O. [1931] Homoiomorphie der kompakten konvexen Mengen im Hilbertschen Raum. Math. Ann., 105, 748758. Kunen, K. a [1978] Weak P -points in . In Topology, Coll. Math. Soc. Bolyai Jnos 23, pages 741749. Budapest (Hungary). Malykhin, V. I. e [1979] is prime. Bull. Acad. Polon. Sci. Sr. Math. Astronom. Phys., 27, 295297. Maurice, M. A. [1964] Compact Ordered Spaces. MC Tracts 6, Mathematical Centre, Amsterdam. van Mill, J. [1984] An Introduction to . In Handbook of Set-Theoretic Topology, K. Kunen and J. Vaughan, editors, chapter 11, pages 503568. North-Holland, Amsterdam.

You might also like