The failure of the uncountable non-commutative Specker Phenomenon

By Saharon Shelah

and Lutz Str ngmann u

Abstract Higman proved in 1952 that every free group is non-commutatively slender, this means that, for a free group G and for a homomorphism h from the free complete product Z of countably many copies of Z into G, there exists a nite subset F and a homomorphism h : F Z G such that h = hF where F is the natural map from Z into F Z. Due to the corresponding phenomenon for abelian groups this is called the noncommutative Specker Phenomenon. In the present paper we shall show that Higmans result fails if one passes from countable to uncountable and with it answer a question posed by K. Eda. In particular, we will see that, for an uncountable cardinal and for non-trivial groups G ( ), there are 22 homomorphisms from the free complete product of the G s into the integers.

modified:2000-12-06

Introduction
In 1952 Higman [H] proved that every free group G is non-commutatively slender where slenderness means that any homomorphism h from the free complete product Z of countably many copies of the integers into G depends on nitely many coordinates only. A similar result was proven by Specker in 1950 [S] for abelian groups. Specker showed that any homomorphism from the product Z of countably many copies of Z into the integers is determined by only nitely many entries. These two phenomenons are called the commutative and
Mathematics Subject Classication (2000): 20E06. 729 in list of publications. Supported by GIF project No. G-0545-173.06/97 of the German-Israeli Foundation for Scientic Research & Development Supported by the Graduiertenkolleg Theoretische und Experimentelle Methoden der Reinen Mathematik of Essen University.
Number

revision:2000-12-06

729

the non-commutative Specker Phenomenon, respectively. Eda [E1] extended Higmans result by showing that, for any non-commutatively slender group S, for any non-trivial groups G ( I) and for any homomorphism h from the free -product of the G s into S, there exist a nite subset F of I and a homomorphism h : iF Gi S such that h = hF where F is the natural map from Gi to iF Gi (for the denition of -product see 1.2). Motivated by iI this result Eda [E1, Question 3.8] asked whether or not the non-commutative Specker Phenomenon still holds if one passes from countable to uncountable cardinals replacing Z by the free complete product Z for some uncountable cardinal (see 1.2). Here we shall give a negative answer to Edas question by constructing, for a given uncountable cardinal and for non-trivial groups G ( ), a homomorphism h from the free complete product of the G s into Z for which the non-commutative Specker Phenomenon fails. In fact, we will show that there are 22 of these homomorphisms and so, in particular, we have that the cardinality of the set of all homomorphisms from G into the additive group of the ring of integers is the largest one possible. This contrasts the countable case and also the abelian case.

Basics and notations

Let I be an arbitrary set. For groups Gi (i I), the free product is denoted by iI Gi (for details on free products see [M]). Given arbitrary subsets X Y of I we put XY : iY Gi iX Gi to be the canonical homomorphism. Moreover, we use the notation X I for nite subsets X of I. Then the set {iX Gi : X I} together with the homomorphisms XY (X Y I) form an inverse system; its inverse limit lim(iX Gi , XY : X Y I) is called the unrestricted free product of the

modified:2000-12-06

Gi s (see [H]). Eda [E1] introduced an innite version of free products and dened the free complete product iI Gi of the groups Gi (for the exact denition see 1.2); it is isomorphic to the subgroup F I {iF Gi lim(iX Gi , XY : X Y I)} of the unrestricted free product. For the convenience of the reader who is not familiar with the notion of a free complete product we recall the denition of words of innite length and also the denition of and some basic facts about iI Gi as can be found in [E1]. Denition 1.1 Let Gi (i I) be non-trivial groups such that Gi Gj = {e} for i = j I. The elements of iI Gi are called letters. A word W is a function W : W iI Gi from a linearly ordered set W into the set of all letters iI Gi such that W 1 (Gi ) is nite for any i I. If the domain W of the word W is countable then we say that W is a -word. 2

729

revision:2000-12-06

The class of all words is denoted by W(Gi : i I) (abbreviated by W) and the class of all -words is denoted by W (Gi : i I) (abbreviated by W ). Two words U and V are said to be isomorphic (U V ) if there exists an order= isomorphism : U V between the linearly ordered sets U and V such that U () = V (()) for all U . Identifying isomorphic words it is easily seen that W is, in fact, a set. Moreover, for words of nite length (i.e. with nite domain) the above denition obviously coincides with the usual denition of words. For a subset X of the set I the restricted word (or subword) WX of W is given Gi } by the function WX : W X iX Gi with W X = { W : W ()
iX

and WX () = W () for all W X . Therefore WX W. Using restricted words with respect to nite subsets of I we dene an equivalence relation on W by saying that two words U and V are equivalent (U V ) if UF = VF for all F I where we may consider UF and VF as elements of the free product iF Gi . The equivalence class of a word W is denoted by [W ] and the composition of two words as well as the inverse of a word are dened naturally. Thus W/ = {[W ] : W W} together with the representative-wise dened composition form a group. Denition 1.2 Given groups Gi (i I) the free complete product iI Gi is dened to be the group W(Gi : i I)/ as described above. Moreover, the free -product Gi is the group W (Gi : i I)/ which is a subgroup of iI iI Gi . If Gi is isomorphic to a xed group G for all i I then we write I G and G I instead of iI Gi and Gi , respectively. iI Note, for a nite set I, we obviously have that iI Gi and Gi are isomor iI phic to iI Gi . In general we have, by [E1, Proposition 1.8], the free complete product iI Gi is isomorphic to the subgroup F I {iF Gi lim(iX Gi , XY :

modified:2000-12-06

XY I)} of the unrestricted free product. Moreover, Eda [E1] proved that each equivalence class [W ] is determined uniquely by a reduced word; a word W W(Gi : i I) is said to be reduced if W U XV implies [X] = e for any = non-empty word X, where e is the identity, and it never occurs that the letters W () and W () belong to the same Gi for neighbouring elements and of W. Lemma 1.3 (Eda, [E1]) For any word W W(Gi : i I) there exists a reduced word V W(Gi : i I) such that [W ] = [V ] and V is unique up to isomorphism. Furthermore, Eda [E1] showed the following lemma where a word W W(Gi : i I) is called quasi-reduced if the reduced word of W can be obtained by 3

729

revision:2000-12-06

multiplying neighbouring elements without cancellation. Lemma 1.4 (Eda, [E1]) For any two reduced words W, V W(Gi : i I) there exist reduced words V1 , W1 , M W(Gi : i I) such that W W1 M , = V M 1 V1 and W1 V1 is quasi-reduced. = We would like to remark that the free -product Z is isomorphic to the I fundamental group (see [E1]) and the free complete product I Z is isomorphic to the big fundamental group of the Hawaiian earring with I-many circles (see [CC]). Hence free complete products are also of topological interest.

The uncountable Specker Phenomenon

In 1950 E. Specker [S] proved that, for any homomorphism h from the direct product Z of countably many copies of Z into the additive group of the ring of integers Z, there exist a nite subset F of and a homomorphism h : ZF Z satisfying h = hF where F : Z ZF is the canonical projection. This result is called the Specker Phenomenon. It can be easily seen that Speckers result still holds if one considers homomorphisms into any free abelian group G instead of homomorphisms into Z, i.e. free abelian groups are slender. In general, an abelian group G is said to be slender if G satises the above property for any homomorphisms h : Z G. For generalizations to products of uncountably many copies of Z within the category of abelian groups we refer to [EM] or [F1]. In [E2] Eda introduced a non-commutative version of slenderness and that is exactly what we shall consider here. Denition 2.1 A group G is non-commutatively slender if, for any homomorphism h : Z G, there exists a natural number n such that h(\{1, ,n} Z) = {e} where e denotes the identity element of G. Eda proved that non-commutatively slender groups are torsion-free and that non-commuative slenderness for abelian groups is the same as the ordinary (commutative) slenderness (see [E1, Theorem 3.3. and Corollary 3.4.]). Moreover, he proved that non-commutatively slender groups have the following nice property: Proposition 2.2 (Eda, [E1]) Let Gi (i I) be non-trivial groups, let S be a non-commutatively slender group, and let h : Gi S be a homomorphism. iI Then there exist a nite subset F of I and a homomorphism h : iF Gi S such that h = hF where F is the canonical map from Gi to iF Gi . iI

revision:2000-12-06

modified:2000-12-06

729

Another interesting result is that the restricted direct product and the free product of non-commutatively slender groups Sj (j J) are non-commutatively slender (see [E1, Theorem 3.6.]). However, the rst fundamental result on the class of non-commutatively slender groups was already obtained by Higman [H] in 1952; it is stated below. Theorem 2.3 (Higman, [H]) Every free group is non-commutatively slender. Contrary to Higmans result above we will show that the non-commutative Specker Phenomenon fails if one replaces the product of countably many groups by products of uncountably many. To be more precise: we actually show that, for an uncountable cardinal , there are 22 homomorphisms from the free complete product G of non-trivial groups G ( ) into the additive group of the ring of integers. To make the proof more transparent we rst construct one homomorphism for which the Specker Phenomenon fails and then modify the construction to obtain our main result. Theorem 2.4 Let be any uncountable cardinal and G ( ) non-trivial groups. Then there exists a homomorphism : G Z for which the Specker Phenomenon fails. Proof. Let G ( ) be a collection of non-trivial groups with identity elements e and choose elements g = e of G for each . For any regular uncountable cardinal we dene the word M G as follows:
modified:2000-12-06

M : (, <)

G via g

where < is the natural ordering of . Note that M is a word of uncountable conality (i.e. the domain of M has uncountable conality) since is regular and uncountable. For < we put M, to be the subword M [,) of M . Now, let X be any reduced word in G and recall that a subset J (X, <) is called convex if x < y < z and x, z J imply y J. We dene Occ+ (X) := {J (X, <) : J is convex and X
J

M, for some < }. =

revision:2000-12-06

Thus Occ+ (X) counts the occurencies of end segments of M in X. Similarly we let Occ (X) := {J (X, <) : J is convex and X
J

M 1 for some < }. = ,

In order to avoid counting subsets of (X, <) more often than necessary we dene the following equivalence relation on Occ+ (X) and Occ (X): 5

729

Dealing with Occ+ two convex subsets J1 , J2 of (X, <) are said to be equivalent (J1 J2 ) if they have a common end segment; in other words J1 J2 if there exist j1 J1 , j2 J2 such that X S1 X S2 where Si = {j Ji : j = ji } (i = 1, 2). Similarly, if we deal with Occ , we dene the equivalence relation substituting end segments by initial segments. For simplicity we denote both equivalence relations by but the reader should keep in mind that is dened dierently for Occ+ and Occ . First we prove that two subsets J1 , J2 Occ+ (X) are either disjoint or equiv alent. To do so assume that J1 , J2 Occ+ (X) are not disjoint and let j J1 J2 (= ). Moreover, there are ordinals 1 , 2 < ( ) and isomorphisms hi : M,i X Ji (i = 1, 2) since J1 , J2 are elements of Occ+ (X). Thus we can nd i i such that hi (i ) = j and therefore X(j ) = gi for i = 1, 2. Hence 1 = 2 and by transnite induction we conclude X T1 X T2 , where = Ti = {j Ji : j j }. Note that hi is an isomorphism of linearly ordered sets and hence hi commutes with limits and the successor-function. Similarly, two subsets J1 , J2 of Occ (X) are either disjoint or equivalent. Next we show that the set Occ+ (X)/ is nite; by similar arguments it then also follows that Occ (X)/ is nite. Let us assume the contrary, that is, there exist innitely many pairwise non-equivalent Jn Occ+ (X) (n ). Then Jn and Jm are disjoint for any n = m by the above. For each n let X Jn M,n for some n < . Thus = n n is strictly less than as = is regular uncountable and hence cf () > 0 . Since [n , ) for all n we can nd jn Jn such that X(jn ) = M,n () = M, ()
modified:2000-12-06

for all n . But all Jn are pairwise disjoint and therefore X 1 (G ) is innite which contradicts the denition of a word (see 1.1). Thus Occ+ (X)/ and also Occ (X)/ are nite sets. We now dene : G Z as follows: W |Occ+ (X)/ | |Occ (X)/ | where X is the reduced word corresponding to W . Note that is well dened by Lemma 1.3. Moreover, it follows immediately from the denition that (X 1 ) = (X) and also the Specker Phenomenon obviously fails for . Note that in general the sets Occ+ (X) and Occ (X) are not of the same size, e.g. (M, ) = 1. It remains to show, however, that is a homomorphism. Therefore let X and Y be reduced words. By Lemma 1.4 there exist reduced words X1 , Y1 and M such that X X1 M and Y M 1 Y1 and X1 Y1 is quasi= = reduced. Now it is easy to check that (XY ) = (X1 Y1 ) by denition and the fact that XY = X1 M M 1 Y1 . Hence (XY ) = (X1 Y1 ) = (X1 ) + (Y1 ) = (X) + (Y ), 6

729

revision:2000-12-06

as X1 Y1 is quasi-reduced and thus the reduced word of X1 Y1 is obtained without cancellation. 2 We would like to remark that the uncountability of in Theorem 2.4 is essential for the denition of the homomorphism and can not be omitted because of Higmans theorem. Modifying the proof of Theorem 2.4 we obtain: Theorem 2.5 Let be any uncountable cardinal and G ( ) be non-trivial groups. Then there are 22 homomorphisms from the free complete product of the G s into the additive group of the ring of integers. In fact there is an epimorphism from G onto the free abelian group of 2 copies of the integers. Proof. Let be uncountable and {G : } be given as stated. We choose the following family of reduced words M for 2 . First we choose non-trivial elements e = g G for . Let {I : 1 } be a family of pairwise disjoint subsets of each of which has cardinality . It is well-known (see e.g. [EK]) that for every 1 we can nd a family {I , I : 2 } of subsets of I such that any nite Boolean combination of them is of cardinality and moreover there is I that belongs to each I , for every 2 . For every 1 and for every 2 we choose a word M , such that its domain M , equals I , and M , () = g for I , . Then the composition M = M , is a well-dened reduced word in W(G :
1

modified:2000-12-06

) for every 2 . Before dening the claimed homomorphism let us rst state the crucial condition satised by the M s ( 2 ): (i) the domain M of M is well-ordered of order type 1 (the ordinal product) for each 2 and therefore has uncountable conality. Now we repeat the construction given in Theorem 2.4 replacing by M and for a reduced word X and 2 we dene Occ+ (X) = {J (X, <) : J convex, X and Occ (X) := {J (X, <) : J convex, X
J= J

M, for some M } =

M 1 for some M }, ,

revision:2000-12-06

where M, is the end segment of M starting with the element , i.e. M, = M {M :} . As in the proof of Theorem 2.4 two equivalence relations are dened on the sets Occ+ (X) and Occ (X) both denoted by . We are now able to dene homomorphisms from the free complete product of the G s to the integers for each 2 . As in the proof of Theorem 2.4 7

729

we can see that the sets Occ+ (X)/ and Occ (X)/ are nite for any reduced word X and 2 . Moreover, the maps : G Z dened by V |Occ+ (X)/ | |Occ (X)/ | where X is the reduced word corresponding to V , are well dened homomorphisms since I , for every 2 . To obtain 22 homomorphisms we will show that there is a surjection onto the free abelian group of 2 copies of the integers. Dene : G
2

via (V )() = (X) for a word V where X is the reduced word corresponding to V . As all the s ( 2 ) are homomorphisms, so is and we claim that is actually a homomorphism from the free complete product of the G s onto the direct sum Z. First assume that this mapping is not into, of 2 copies of the integers
2

then there is a reduced word X and a sequence of pairwise distinct ordinals n (n ) such that n (X) = 0 for all n . Thus for each n there is a convex subset Jn X such that w.l.o.g. X
Jn =

n ,n

modified:2000-12-06

for some n M n ,n M n ( n 1 ). But now, if 1 such that n < for all n , then the element g (since belongs to all sets I,n ) appears 1 innitely many times in X, i.e. X (G ) is innite - a contradiction. Thus the image of is contained in the direct sum of 2 copies of the integers. On the other hand, by the choice of the sets I , we certainly have that Occ+ (M ) = 0 = Occ (M ) for distinct , 2 . Moreover, Occ+ (M ) = 1 and Occ (M ) = 0 for any 2 . Thus we obtain (M ) = (0, , 0, 1 , 0, )
2

revision:2000-12-06

and therefore is obviously surjective. Since there are 22 homomorphisms from the direct sum of copies of Z to Z itself we are done. 2 We have a short remark. 8

729

Remark 2.6 Note that the above proof gives us that (i) the free complete product G = G contains a free subgroup H (the group generated by the words M ) and; (ii) there is a projection onto H. The following theorem gives us a complete description of all interesting homomorphisms from G = G to the integers for uncountable and groups G ( ). By interesting we mean interesting with respect to the Specker Phenomenon, i.e. if W is a nite subset of , then all homomorphisms from the subproduct GW = W G to the integers extend naturally to a homomorphism from G to Z but these homomorphisms are not of particular interest for us and well-understood, hence we will restrict ourselves to homomorphisms from G to Z which are already zero on every nite subproduct of G. First note that the denition of M in the proof of Theorem 2.5 did not really depend on the particular word M but only on the fact that M had uncountable conality. It is immediate to see that for any word M such that the domain M has uncountable conality we can dene such a homomorphism M : G Z. Hence we dene the following set: IG = {M G : cf (M ) 1 } and let G : G
M IG

ZM

modified:2000-12-06

be dened by G (V ) = (M (X) : M IG ) where X is the reduced word corresponding to V . Now G is well-dened and we have the following theorem. Theorem 2.7 Let G = G for some uncountable cardinal and groups G ( ). Then any homomorphism : G Z which is zero on every nite subproduct of G factors through G . Proof. For simplicity we assume that all words are already in reduced form. First we will show that the kernel of G is exactly Ker(G ) = {M G : M contains no monotonic sequence of length 1 }, where a monotonic sequence of length 1 is just a subset of M which is isomorhic to 1 or its inverse. Clearly we have that Ker(G ) is contained in {M G : M contains no monotonic sequence of length 1 }. Conversely, if M is a reduced word that contains no monotonic sequence of length 1 and G (M ) = 0, then there exists a reduced word N IG , i.e. N has uncountable conality, such that N (M ) = 0. But then Occ+ (M ) or Occ (M ) is non-trivial and hence M must contain a N N monotonic sequence of length 1 - a contradiction. It is now enough to prove that any homomorphism f : G Z which is zero on any nite subproduct of 9

729

revision:2000-12-06

G acts trivial on Ker(G ). For this assume that M is a reduced word which contains no monotonic sequence of length 1 such that f (M ) = 0 for some homomorphism f : G Z. We distinguish between three cases: Case a: There exist subwords Nn of M (n ) such that (i) N n is a convex subset of M ; (ii) the N n s (n ) are pairwise (almost) disjoint; (iii) f (Nn ) = 0 (without loss of generality f (Nn ) > 0). Hence the composition N of the words Nn (n ) is a well-dened word in G and applying Theorem 2.3 together with [E1, Proposition 1.9] leads to a contradiction. Case b: There is an initial segment M of M such that (i) f (M ) = 0, where M = M
M

(ii) for every proper initial segment N of M N = M N;

we have f (N ) = 0, where

(iii) M has no largest element or for every t M there exists a convex subset N t {m M : m t} such that f (Nt ) = 0, where Nt = M N t . Then the conality of M has to be 0 by the assumptions and we choose an increasing, unbounded sequence {tn : n } in M and put Nn = {m M : m tn } or Nn = Ntn (hence f (Nn ) = 0). In both cases we easily obtain a contradiction. Similarly we can use the same arguments for the inverse of M to get a contradiction. Case c: Neither case a nor case b is satised. Then it is easy to see that the set J = {t M : f (Mt ) = 0} is nite, where Mt = M {t} (e.g. use Ramseys Theorem). So without loss of generality we may assume that J is the empty set. We let

revision:2000-12-06

modified:2000-12-06

I = {A M : A convex and B A convex f (M

B)

= 0}.

Then I contains all singletons t M, the empty set and it is downwards closed. Moreover, if A and B are elements of I and A B is convex, then A B I. Finally every initial segment of M with no largest element has an end segment 10

729

in I and hence M I - a contradiction. Similarly we obtain a contradiction using the inverse M 1 instead of M . This nishes the proof. 2 For completeness let us state the following remark. Remark 2.8 If h : G Z is any homomorphism, then an application of Theorem 2.3 shows that the set { : h(G ) = {0}} = F is nite. Hence, regarding F G as a subgroup of G we let h0 = (h F G )F . Then h h0 satises the assumptions of Theorem 2.7 and thus factors through G . Acknowledgement: The authors would like to thank the referee for his helpful comments and suggestions and in particular for his patience.

References
[CC] J. W. Cannon and G. Conner, The big fundamental group, big Hawaiian earrings, and the big free groups, Topology Appl. 106 (2000), 273291. [CE] G. Conner and K. Eda, Free subgroups of complete free products, preprint (2000). [E1] K. Eda, Free -products and noncommutatively slender groups, J. of Algebra 148 (1992), 243263. [E2] K. Eda, The non-commutative Specker Phenomenon, J. of Algebra 204 (1998), 95107. [EK] R. Engelking and M. Karlowicz, Some theorems of set theory and their topological consequences, Fund. Math. 57 (1965), 275285. [EM] P.C. Eklof and A. Mekler, Almost Free Modules, Set-Theoretic Methods, North-Holland Amsterdam - New York (1990). [F1] L. Fuchs, Innite Abelian Groups - Volume I and II, Academic Press, New York - London (1970 and 1973). [F2] L. Fuchs, Abelian Groups, Hungarian Academy of Science, Budapest (1958).

modified:2000-12-06

revision:2000-12-06

[H] G. Higman, Unrestricted free products and varieties of topological groups, J. London Math. Soc. 27 (1952), 7381. [M] W.S. Massey, Algebraic Topology: An Introduction, Graduate Texts in Mathematics, Springer Verlag New York - Heidelberg - Berlin (1967).

11

729

[Sh] S. Shelah, Classication Theory, Studies in Logic and The Foundations of Mathematics, NorthHolland Amsterdam - New York - Oxford - Tokyo 92 (1990). [S] E. Specker, Additive Gruppen von Folgen ganzer Zahlen, Portugal. Math. 9 (1950), 131140. Saharon Shelah Department of Mathematics Hebrew University Jerusalem, Israel and Rutgers University Newbrunswick, NJ U.S.A. e-Mail: Shelah@math.huji.ac.il Lutz Strngmann u Fachbereich 6, Mathematik University of Essen 45117 Essen Germany e-Mail: lutz.struengmann@uni-essen.de

revision:2000-12-06

modified:2000-12-06

12

729