Ordered Groups and Infinite Permutation Groups

Mathematics and Its Applications

Managing Editor:

M. HAZEWINKEL
Centre for Mathematics and Computer Science. Amsterdam. The Netherlands

Volume 354
Ordered Groups and
Infinite Permutation
Groups

edited by

W. Charles Holland
Bowling Green State University

KLUWER ACADEMIC PUBLISHERS

DORDRECHT / BOSTON / LONDON
A c.I.P. Catalogue record for this book is available from the Library of Congress.

ISBN-13: 978-1-4613-3445-3 e-ISBN-13: 978-1-4613-3443-9

001: 10.1007/978-1-4613-3443-9

Published by Kluwer Academic Publishers,

P.O. Box 17, 3300 AA Dordrecht, The Netherlands.

Kluwer Academic Publishers incorporates

the publishing programmes of
D. Reidel, Martinus Nijhoff, Dr W. Junk and MTP Press.

Sold and distributed in the U.S.A. and Canada

by Kluwer Academic Publishers,
101 Philip Drive, Norwell, MA 02061, U.S.A.

In all other countries, sold and distributed

by Kluwer Academic Publishers Group,
P.O. Box 322, 3300 AH Dordrecht, The Netherlands.

Printed on acid-free paper

All Rights Reserved

© 1996 Kluwer Academic Publishers
No part of the material protected by this copyright notice may be reproduced or
utilized in any form or by any means, electronic or mechanical,
including photocopying, recording or by any information storage and
retrieval system, without written permission from the copyright owner.
Contents
Preface ...................................................................... vii
V. M. KOPYTOV AND N. YA. MEDVEDEV
Quasivarieties and Varieties of Lattice- Ordered Groups . ........................ 1
S. H. MCCLEARY
Lattice-ordered Permutation Groups: The Structure Theory ... ................. 29
J. K. TRUSS
On Recovering Structures from Quotients of their Automorphism Groups . ..... 63
M. DROSTE AND R. GOBEL
The Automorphism Groups of Generalized McLain Groups ................... 97
M. RUBIN
Locally Moving Groups and Reconstruction Problems ........................ 121
S. A. ADELEKE
Infinite Jordan Permutation Groups ........................................ 159
C. E. PRAEGER
The Separation Theorem for Group Actions ................................. 195
D. MACPHERSON
Permutation Groups Whose Subgroups Have Just Finitely Many Orbits ... ... 221
J. L. ALPERIN, J. COVINGTON, AND D. MACPHERSON
Automorphisms of Quotients of Symmetric Groups ........ .................. 231

v
Preface

The subjects of ordered groups and of infinite permutation groups have long en-
joyed a symbiotic relationship. Although the two subjects come from very different
sources, they have in certain ways come together, and each has derived considerable
benefit from the other.
My own personal contact with this interaction began in 1961. I had done Ph. D.
work on sequence convergence in totally ordered groups under the direction of Paul
Conrad. In the process, I had encountered "pseudo-convergent" sequences in an
ordered group G, which are like Cauchy sequences, except that the differences be-
tween terms of large index approach not 0 but a convex subgroup G of G. If G
is normal, then such sequences are conveniently described as Cauchy sequences in
the quotient ordered group GIG. If G is not normal, of course GIG has no group
structure, though it is still a totally ordered set. The best that can be said is that
the elements of G permute GIG in an order-preserving fashion. In independent
investigations around that time, both P. Conrad and P. Cohn had showed that a
group admits a total right ordering if and only if the group is a group of automor-
phisms of a totally ordered set. (In a right ordered group, the order is required
to be preserved by all right translations, unlike a (two-sided) ordered group, where
both right and left translations must preserve the order.)
It was at about that time that I had the good fortune to be able to attend the
lectures of Helmut Wielandt and Olaf Tamaschke in Tiibingen. Wielandt's lecture
notes on infinite permutation groups had just become available, and I became aware
of the powerful methods of that theory. In particular, the most natural context for
my problem on Cauchy sequences is that of a group acting on a totally ordered set.
Such a group is not itself necessarily totally ordered (two-sided), but is (two-sided)
lattice ordered by the naturally inherited pointwise order. Indeed, G. Birkhoff had
observed, in his book on lattice theory, that the group of automorphisms of a totally
ordered set is a natural example of a (usually, non-Abelian) lattice-ordered group,
and F. Sik had investigated the structure of such automorphism groups. Using
ideas based on my Cauchy sequence problem, I was able to show that every lattice-
ordered group is embeddable in the lattice-ordered group of automorphisms of a
totally ordered set. I then found the obvious analogues in the context of lattice-
ordered groups for many of the basic results about general permutation groups.

VII
VUl

Since then, these methods have become crucial in the study of non-Abelian lattice-
ordered groups. From the other direction, it was already pointed out in Wielandt's
lecture notes that certain permutation groups (the "primitive but not strongly prim-
itive" ones) can be viewed as automorphism groups of a certain partial order on
the permuted set. The pursuit of similar ideas has led a portion of the theory of
permutation groups in the direction of viewing them as automorphism groups of
certain structures on the permuted set.
In the summer of 1993, a conference at Luminy in France, organized by M. Gi-
raudet, F. Lucas, and D. Gluschankof, provided the opportunity for many experts
in ordered groups and in infinite permutation groups to explore the areas of overlap
between the two subjects. The present volume is not, in the usual sense, a proceed-
ings of that conference; not all of the authors represented here were participants
at the conference, and not all participants are represented. Rather, I have tried to
choose a sample of articles by various authorities to represent a spectrum of ideas
ranging from almost purely ordered groups at one end to almost purely permutation
groups at the other. I hope that in this way the reader can get a good feeling for
the interaction of the two subjects.
The chapter in the present volume by V. M. Kopytov and N. Ya. Medvedev
on quasi varieties of lattice-ordered groups, while mainly of a purely algebraic na-
ture, still contains many results whose proofs rely on the representation as ordered
permutation groups.
The structure theory of lattice-ordered permutation groups, much of it due to
S. H. McCleary, is surveyed in McCleary's chapter in the present volume.
In the chapter in this volume on Jordan groups by S. A. Adeleke, the permutation
groups become automorphism groups of various order-like structures.
From another standpoint, if G is the group of all automorphisms of some struc-
ture, then certain properties of the structure are reflected in G itself. In the best
cases, one can "reconstruct" the structure from G. This is the subject of the chap-
ters in this volume by J. K. Truss (the structure is a totally ordered set and the
group of interest is a certain quotient of the full automorphism group), M. Droste
and R. Gobel (the structure is a McLain group), and especially the very general ap-
proach in the chapter of M. Rubin; to some extent, the same ideas are present also
in the chapter by J. L. Alperin, J. Covington, and D. Macpherson (the structure is
a quotient of an infinite symmetric group).
The chapters of C. Praeger and of D. Macpherson are essentially of a permutation-
group-theoretic nature, but still use methods that are strongly reminiscent of those
used in the more order-theoretic parts of the subject.
To all of these authors, to the three organizers of the Luminy conference, and
to several anonymous referees, I express my sincerest gratitude.

W. Charles Holland
In the Black Swamp
August, 1995
 Quasivarieties and Varieties
of Lattice-Ordered Groups *

V. M. Kopytov N. Va. Medvedev

Mathematical Institute Altai State University

Siberian Academy of Science Barnaul 656099
Novosibirsk 630090 Russia
Russia

1 Introduction
Many properties and statements of the theory of lattice-ordered groups (i-groups)
can be formulated and proved in terms of first order logic. Special mention should
be made of properties expressed by universal sentences such as identities and im-
plications, which can be referred to as the theory of varieties and quasivarieties,
respectively, of i-groups. The theory of varieties of i-groups has been developed
for more than two decades and contributions to it have been included in books and
survey articles (Anderson and Feil [1], Kopytov and Medvedev [44, 45], Reily [63]).
It was not until the mid-80s that the systematic investigation of the theory of qua-
sivarieties of i-groups began and the results obtained in this area are not available
for wide audience yet. It is clear that the theory of quasi varieties is more general
than that of varieties. Nevertheless, there have been obtained a number of results
on quasivarieties of i-groups asserting that helpful and non-trivial properties of i-
groups can be defined by means of implications, and the theory of quasivarieties of
i-groups itself is exciting and profound.
This paper consists of two parts which are different in style and goal. The first
part (sections 2 through 6) contains an introduction to the theory of quasivarieties
of i-groups and presents the statements and proofs of those results whieh are, in
the authors' opinion, the most interesting. The second part (sections 7 through 11)
is a survey of results in the theory of i-varieties obtained after the publication of
N. Reilly's paper [63](1989). The account in the present paper is "semi closed" since
we assume the standard definitions and results on i-groups from the well-known

·This work was done with financial support of the Russian Fund of Fundamental Research
(project code 93-011-1524)

W. C. Holland (ed.), Ordered Groups and Infinite Permutation Groups, 1-28.

© 1996 Kluwer Academic Publishers.
2 Kopytov and Medvedev

books and survey articles, while the notions not previously defined are explained in
the article. We employ standard notation and terminology of group theory (Hall
[28], Kargapolov and Merzljakov [38]) and £-group theory (Reily [63]).

PART 1: €-Quasivarieties

2 Generalities
Let us recall that a formula <p of t.he first-order language and of the signature (J' is
an implication (or quasi-identity) if <p is of the form
(VXl, ... , Xn)((Al = B l &A 2 = B2&" . &Ak = B k) => (A = B))
where AI,,'" A k , B l , ... , Ih, A, B are terms of signature in free variables
(J'

Xl, ... , X n . The class K. of algebraic systems of signature is a quasivariety if

(J'

there is a set. cl> of implications such that an algebraic system G belongs to K. iff all
implications of cl> are valid in G.
An implication of signature £ = {. , -1, e , V, I\} is a formula <p of the predicate
calculus of the form
(VX!) ... (VXn)((Wl(Xl, ... , xn) = e& ... &Wk(Xl, ... , xn) = e) =>
=> Wk+l(Xl, ... , xn) = e),
where Wl(Xl, ... , xn), ... , Wk+l(Xl, ... , xn) are i-group words. An i-group G satis-
fies the implication <p if whenever

then Wk+l(gl, ... ,gn) = e.

A quasivariety of i-groups (or, simply, £-quasivariety) is the class X of all £-
groups which satisfy a given set cl> of implications of the signature £ = {-, -1 , e, V, I\}.
The set cl> is called a basis of the implications of the £-quasivariety X. From standard
results in universal algebra (see Burris and Sankappanavar [7], Theorem 2.25.) we
have the following characterization of i-quasi varieties:
Theorem 2.1 Let .1:' be a class of i-groups. Then the following statements are
equivalent:
(1) .1:' is a quasivariety of £-groups,
(2) X 1S dosed under i-isomorphisms, i-subgroups and reduced products,
and contains a trivial i-group,
(3) X is closed under £-isomorph1sms, i-subgroups, cartesian products and
uitraproducts, and contazns a trivwl i-group.
Quasivarieties and Varieties of Lattice-Ordered Groups 3

It is clear that the intersection of £-quasivarieties is an £-quasivariety. Hence,

given any class K of i-groups, there is a unique minimal £-quasivariety that contains
K. This £-quasivariety is called the £-quasivariety generated by K and is denoted
by qf(K).
For any class K of i-groups let P(K) , S(K) and PuCK) denote the classes of
i-groups which are cartesian products, i-subgroups and ultraproducts, respectively,
of elements of K .
From Theorem 2.1 the following useful observation follows.

Theorem 2.2 If K is any class of i-groups, then the £-quasivariety qf (K) generated
by K is SP PuCK).

The following result can be proved by direct verification.

Proposition 2.3 The conjunction of a finite number of implications in the signa-

ture £ = { . , -1 , e , V, t\} is equivalent to a single implication in the class of
i-groups £; namely, if we denote Igl = g V g-l and x = (Xl, . .. , x n ), then

is equivalent to

VIw!;(x)1 = e)
m

Vx(( -+ w(x) = e).

k=l

Corollary 2.4 If an £-quasivariety X has a finite basis of implications, then X can

be defined by one implication.

It is easy to see that many properties of i-groups can be formulated as impli-

cations in the signature £ = {- , -1 , C , V, t\}. Each i-variety is an £-quasivariety,
but there are £-quasivarieties which are not i-varieties. One such £-quasivariety
is the class of all orderable i-groups. That this is an £-quasivariety follows from
Theorem 2.1. To see that it is not an i-variety, we observe the following. Let DS
be the i-group of all bounded piecewise linear order-preserving permutations of the
totally ordered set R of real numbers.
Then Dr is o-2-transitive on Rand orderable. Thus, D? generates the /-variety £
of all i-groups (Holland [32](1985)). If the orderable i-groups formed an i-variety,
this would imply that all i-groups are orderable. But this is not true since, for
example, the i-group A(R) of all order permutations of the real numbers is not
orderable, because there exist non-trivial I, 9 E A( R) such that 1- 1 9 I := g-l. So
the set A of all £-quasivarieties properly contains the set L of all i-varieties. The set
A of all £-quasivarieties is a lattice where, for any l'-quasivarieties X, y, the greatest
lower bound, or meet, of X and Y is their set-theoretical intersection:
4 Kopytov and Medvedev

and their least upper bound, or join, is the intersection of all i-quasi varieties con-
taining both X and Y:
A

X VY:::: n{W : Xu Y ~ W}.

L
Let X and Y be varieties of i-groups and X V Y be the join of X and Y in the
lattice L of i-varieties. Then by (Martinez [48)(1974)),

if and only if there exist i-ideals M and N in G such that M n N :::: E and
G / M EX, G / N E y. In this case, therefore, G is isomorphic to some i-subgroup
A
of the direct product G / M x G / N and so G E X Vy. Hence, the lattice L of all
i-varieties is a sublattice of the lattice A of all i-quasivarieties.
The lattice A of i-quasivarieties is more diverse and complicated than the lattice
L of i-varieties.
For a discussion of the relationship between free i-groups and and l'-quasivarieties,
see V. M. Kopytov [42)(1979).

3 The lattice of f-quasivarieties

The following result establishes a difference between the lattice L of l'-varieties and
the lattice A of l'-quasivarieties.

Theorem 3.1 (S. A. Gurchenkov [private communication)) The lattice A of l'-

quasivarieties is not modular and therefore is not distributive.

Proof. Consider the group

G :::: gp{a, b I [a bj , a b') :::: [a, b, b, b] :::: e, (i, j E Z)).

It is clear that G is metabelian and torsion-free nilpotent of class::; 3. Now

we define on G two total orders. The total order PI is uniquely defined by the
inequalities b ~ a ~ [a, b] ~ [a, b, b] > e. Let G l be the group G totally ordered
by Pl. The total order P2 is uniquely defined by the inequalities: b ~ a ~ [a, b] ~
[a, b, b]-l > e. The group G totally ordered by P2 is denoted by G2. Let G 3 be the
i-subgroup of the direct product G 1 x G 2 generated by the elements b :::: (b, b), a::::
(a, a). Now consider the following implications (we omit quantifiers):

[x, y] 2: e =} [x, y, y] 2: e;
 Quasivarieties and Varieties of Lattice-Ordered Groups 5

[X,y] ~ e:::} [x,y,y]::; e;

[x, y] ~ e :::} [x, Y, y] = e;

[X,y,y] ~ e:::} [x,y,y] = e;

t3,2: [x, y, y] ::; e :::} [x, y, y] = e,

where x = Izl /\ Iwl, y = Izl V Iwl·
Direct verification shows that 0 1 satisfies tl but not t2, while G 2 satisfies t2 but
not t i . Moreover, 0 3 satisfies both t 3,1 and ts,2 but neither tl nor t2' Let.M be the
i-quasivariety of nilpotent i-groups of class::; 3 defined by the implication t4, and
A A A A
Ml = ql(Gd VM, M2 = ql(G2) VM, M3 = ql(Od VM Vql(03).
Then M l , M 2, M3 are different. Direct verification shows that M2 # Ml V M2 =
M2 V M3 # M3 and Ml V M2 = M. We claim that M3/\M2 = M. Let
H E M3 /\ M2 and z, w E H such that [x, y] ~ e. As H E M 2, then [x, y, y] ::; e.
Since H E M 3, then H is a sub direct product of i-groups HI E M, H2 E ql(Od,
and H3 E q/(03). Then
[x,y,y] = [Xl,YI,yd· [X2,Y2,Y2]' [X3,Y3,Y3],
where [Xj,Yi,Yi] E Hi (i = 1,2,3). Since [x,y] ~ e and [x,y,y] ::; e in H, then
[Xi, y;] ~ e, [Xi, Yi, y;] ::; e in Hi. Since the implications t 3,1, t 3,2 are valid in H 3, the
inequalities are valid in H 3, and the inequalities [X3, Y3] ~ e, and [X3, Y3, Y3] ::; e
imply [X3, Y3, Y3] = e. Since the implication tl holds in H 2, then the inequality
[X2, Y2] ~ e implies [X2, Y2, Y2] = e. Finally, the validity of the implication t4 and the
inequality [Xl, YI] ~ e implies [Xl, Yl, YI] = e. Hence the sub lattice of the lattice A of
i-quasivarieties generated by the i-quasivarieties M 1, M 2, and M3 is a 5-element
non-modular lattice.
It is also known that the sublattice of all those i-quasi varieties which can be
defined by implications of the signature of the group theory { . , -1 , e } is not
modular (N. Ya. Medvedev [51](1984)).
It is well known that the i-variety N of normal-valued i-groups is the greatest
element in the lattice L of i-varieties (Holland [31](1976)). In the lattice A of
i-quasi varieties this is not true, as we see in the following.
For each positive integer n, let Rn be the i-quasivariety defined by the implica-
tion (xn = yn) :::} (x = y). Let 'R... denote the i-quasi variety of all i-groups which
admit unique extraction of all roots, i.e.,
00

'R... = n 'R.. n .
n=2
6 Kopytov and Medvedev

It is clear that the i-quasi variety 0 of all orderable i-groups is contained in R •. As

usual, we denote the Scrimger i-groups as G n (see, for example, Reilly [63]).

LeIllIlla 3.2 (Arora [3](1985)) G n E Rm if and only if (m, n) =1 where (m, n)

denotes the greatest common divisor of m and n.

Corollary 3.3 If P and q are distinct prime numbers, then G p E Rq \ Rp and

G q E Rp \ R q. Hence, Rp and Rq are incomparable i-quasivarieties if P and q are
distinct prime numbers.

Let n be a positive integer;:::: 2 and n = p~' ... p~k where PI, ... Pk are distinct
prime divisors of nand nl, ... ,nk > O. Then an easy induction on k shows that
Rn = R p, n ... n R Pk . By the above arguments, we conclude

Corollary 3.4 Let m and n be positive integers. Then Rn ;2 Rm if and only if

every prime divisor of m is a prime divisor of n.

Let P be the set of positive prime integers and let pi ~ P be any non-empty
subset of P. We define a quasivariety (for each Pi) as follows:

Let PI and P 2 be two incomparable subsets of P; that is, PI R. P 2 and P 2 R. Pl.

Then, Rp, and Rp, are two incomparable i-quasivarieties. In fact, for P E PI \ P2
and q E P 2 \ PI we have G p E Rp2 \ Rp" and G q E Rp, \ Rp,.
Since there is a continuum of pairwise incomparable subsets of P, we have
exhibited a collection of continuum many pairwise incomparable i-quasivarieties.
As before, letDP be the i-group of all bounded piecewise linear order permuta-
tions of the real numbers R. Let H = DpWr(Z, Z). For each pEP, define Hp as
follows:
Hp = gp(({gn},m) E H I gi = gj ifi =j (mod p)).
Since DP
is o-2-transitive, it follows that DP is not normal-valued. Thus, for all
pEP, Hp is not normal-valued either.

LeIllIlla 3.5 (Arora [3](1985)) Hp E Rq \Rp ifp and q are different prime numbers.

LeIllIlla 3.6 Let {Hij liE I, 1 ~ j ~ n < oo} be a collection of i-groups. Let 1)
be an ultrafilter over I. Then

iEI iEI iEI iEI

LeIllIlla 3.7 Let Xj (j E J) be a finite collection of i-quasivarieties and let G be an

i-group. Then G E VjE} Xj if and only if there exist Mj E Xj (j E J) such that G
can be i-embedded in the i-group DjE} Mj.
Quasivarieties and Varieties of Lattice-Ordered Groups 7

A
Lemma 3.8 Hp f/:. NVR p .
Theorem 3.9 (Arora [3](1985)) There exists a continuum of incomparabll:
i-quasivarieties, each properly containing the i-variety N of normal-valued i-groups.

Lemma 3.10 (Arora [3](1985)) Let G be a finitely subdirectly irreducible i-group.

A
Let {Xj Ij E J} be a finite collection of i-quasivarieties. Then G E V Xj if and
jEJ
only if G E Xjo for some jo E J.

Lemma 3.11 (Arora [3](1985)) ql(A(R)) = C.

Theorem 3.12 (Arora [3](1985)) The i-variety C of all i-groups is finitely join-
irreducible in the lattice A of i-quasivarieties.

4 The universal theory of abelian i-groups

Let us recall that a first-order formula <l> is a universal formula if it is in prenex form
and all the quantifiers are universal. An i-group G 1 is universally equivalent to an
i-group G 2 iff the sets Th'v'Gl and Th'v'G2 of universal sentences of the signature
1= {- , -1 , e , V, I\} which are satisfied by G 1 and G 2 , respectively, are the same
(Burris and Sankappanavar [7], p.215). For abelian groups we use the additive
notation.
Theorem 4.1 (Gurevich and Kokorin [27](1963)) All nontrivial totally ordered
abelian groups are universally equivalent.

Theorem 4.2 (Khisamiev and Kokorin [40](1966), Khisamiev [39](1966))

(1) Let G and G' be abelian i-groups of finite orthogonal ranks rand r', respec-
tively. Then G is universally equivalent to G ' if and only if r = r'o
(2) Let G and G ' be abelian i-groups of infinite orthogonal rank. Then G and
G ' are universally equivalent.

It is clear that an i-group of finite orthogonal rank is not universally equivalent

to any i-group of infinite orthogonal rank.
Corollary 4.3 The f-variety A of abelian f-groups is the smallest non-trivial f-
quasivariety in the lattice of i-quasivarieties.

5 Covers of A in the lattice of i-quasi varieties

The structure of the lattice A of i-quasivarieties is much more complicated than
that of the lattice L of i-varieties. For example, the list of covers of A in A is more
diverse than the same in L. As usual we denote by Gp the Scrimger i-group for any
prime number p, and by W+, W-, No the Medvedev i-groups (see Reilly [63]).
8 Kopytov and Medvedev

Proposition 5.1 The i-variety vart(Gp) = Sp = qt(Gp) covers A in the lattzce of

i- quasivarieties.

The proof of Proposition 5.1 follows from the work of Gurchenkov [16](1984),
and Holland, Mekler and Reilly [34](1986).

Proposition 5.2 The i-quasivarieties ql(W+) and ql(W-) cover A in the lattice
of i-quasivarieties.

The proof of Proposition 5.2 follows immediately from Medvedev [49](1977).

By Proposition 5.1 the i-variety Sp = vartCGp) generated by the Scrimger i-
group Gp coincides with the i-quasivariety ql(Gp ) generated by Gp . Therefore, we
have the following classification of the covers of A in the lattice of i-quasivarieties.

Theorem 5.3 (Gurchenkov and Kopytov [26](1987), Van Rie [64](1991)) If X is a

cover of A in the lattice of i-quasivarieties, then one of the following is true:
(1) X Sp= =
ql (Gp) for some prime p where G p is a Scrimger I-group;
(2) X ~ ql(Zwr(Z, Z));
(3) X ~ R.

We now consider nilpotent i-quasivarieties. Let]{o be the group having the

presentation

]{o = gp(a, b, c, c+, c- I c = [a, b] = c+c-, e = [c+, c-] =

= [c+,a] = [c-,a] = [c-,b] = [c+,b])
and lattice-ordered by the rule: ak bl (c+ r (c)n 2: e iff k > 0 or k = 0 and I >0
or k = 1= 0 and m 2: 0, n::; o.

Theorem 5.4 (Medvedev [52](1985)) Each non-abelian nilpotent i-quasivariety con-

tains /{o. Hence, the i-quasivariety qt(I{o) is the unique cover of A in the lattice
of nilpotent i-quasivarieties.

Recalling that No is the nilpotent Medvedev o-group which generates the unique
nilpotent i-variety covering A in the lattice of i-varieties, it is natural to ask whether
qt(No) and ql(I{o) are equal. The answer is negative since the implication

([x,y] Ve = [x,y]):=;. ([x,y] = e)

holds in /{o and violated in the o-group No. Moreover, the representable i-quasivar-
iety ql(/{o) does not contain any non-abelian o-group.
Since the i-variety A is defined by a single identity, by the compactness theorem
(see Burris and Sankappanavar [7]), it follows that each member of A in the lattice
L of varieties of i-groups which properly contains A contains a cover of A in the
lattice A (see, for example, Reilly [63]). All these covers of A are very close to those
Quasivarieties and Varieties of Lattice-Ordered Groups 9

of A in A . Now, however, we will construct an infinite set of covers of A in the

lattice A which are very far from the above covers.
Let G = Zwr(Z, Z) and a = ({ad, 0), b = ({ad, 1) where at = 0 E Z for all
t= = =
t E Z and at = 0 for t 0, ao 1. Then gp(a, b) G and the lower central series

G = 71(G) ~ 72(G) ~ .. ·7n(G) ~ ...

has the following properties (Kargapolov and Merzljakov [38]):

(4) n 7n(G) E;=
nEN
(5) the quotient group 7i+1(G)hi+2(G) = ([a,8 7i +2(G» is an infi-
i
nite cyclic group generated by the element [ a , 8 7 i+2(G).
i

Now for each integer n ~ 2 and each finite set {co, c1, ... , cn _ d, where co =
+1,c/ = ±1(l = 1,2, ... ,n-l), let Qn(cO,c1, ... ,cn-1) be the total order ofG =
Zwr(Z, Z) uniquely defined by the following inequalities:
(6) b :;p. a :;p. Ha, bJi :;p. I[a, b, b]1 :;p. .. ·I[a, b, ... , bJi ~ ... > e;
...............
k
(7) [a, b, ... , b]<;; > e;
...............
k
where k == k (mod n) and k E {O, 1, . .. ,n-l}.
Direct verification shows that if [x, y] t=
e in the o-group

for some x, y E G, then Ixl V Iyl = bStp, where s > 0 and tp belongs to the abelian
radical A(G) (the largest abelian convex i-subgroup). Let

v'C) (x,y,cO,Cl, .. . ,cn-d

(u~i)(x, yyD V e) /\ (uii)(x, yY') /\ ... /\ (U~~l(X, Y)~-l Ve),
w(x, y) w(x, y, co, ... , cn-d
n-1
V v\x, y, co,··., cn-d,
;=0
w(x, y) I[w(x, y), Ixl V IYIJI,

where

u~O)(x, y) I[x, y]l,

u~i)(x, y) = Hu~i-l)(x, y), Ixl V IYIl/' i = 1,2, ... , n - 1;
uji)(x, y) [u?2 1 (x, y), Ixl V Iyl]' j = 1,2, ... , n - 1.
10 Kopytov and Medvedev

Let fa = +l,f; = -1, i E {1,2, ... ,n - I} and let (G,Qn) be the group
G = Zwr(Z, Z) totally ordered by the order

Qn = Qn(l, -1, -1, ... , -1).

Lemma 5.5 Let x, y E G and [x, y] f. e. Then

w(x, y) = [a, --..--

b, ... , Wip' > e in (G, Qn),
nt-1

where rp' E Int+1 (G), and l' is a negative integer.

Lemma 5.6 Let x, y E G and [x, y] i= e. Then the totally ordered subgroup H =
gp(w(x, y), Ixl V Iyl) of (G, Qn) generated by the elements w(x, y) and Ixl V Iyl is
order isomorphic to the o-group (G, Qn).

Lemma 5.7 Let V = (G, Qn l I:F be an ultrapower and x, y E V such that [x, y] i=
e. Then the subgroup H = gp(w(x, y), Ixl V Iyl) of V is order isomorphic to the
o-group (G, Qn).

Lemma 5.8 Let x, y E DjE ] Vj, where Vi = (0, Qn)l] IF is an ultrapower of the 0-
group (0, Qn) for each j E J and [x, y] i= c. Then the subgroup H = gp(w(x, y), Ixlv
Iyl) ofD jE ] Vi is totally ordered and IS order isomorphic to (G, Qn).
The proofs of Lemmas 2, 3 and 1 are all similar.
By Theorem 2.2, gl(G, Qn) = SP Pu(G, Qn), where S, P, Pu are the operations
of taking £-subgroups, cartesian products and ultraproducts, respectively. Let H be
any non-abelian £-group and H E gl(G, Qn). Then by Lemma 5.8, H contains an
o-subgroup which is order isomorphic to (G, Qn). This yields the following result.

Theorem 5.9 (Isaeva and Medvedev [36](1992)) Each £-quasivariety

gl(G,Qn), n E Z,n ~ 2

has the following pT'operties:

(8) gl(G, Qn) covers A in the lattice A of £-quasivaneties,
(9) gl(G, Qn) is contained in the I-variety Wa of weakly abelian I-groups,
(10) each nilpotent f-group of ql(G, Qn) is abelian.

It is evident that ql(G, Qm) f. gl(G, Qn) if n f. m. In fact, let us assume,

on the contrary, that gl(G, Qm) = ge(G, Qn) for m < n. From the definition
of the total orders Qn and Qm and Lemma ,').5 it follows that the implication
«w(x, y) = e) =? ([x, y] = e)) holds in (G, Qn) and is violated in (0, Qm) if x = a
and y = b.
Here we construct a new cover of A in the lattice A with the following properties:
 Quasivarieties and Varieties of Lattice-Ordered Groups 11

(11) it does not contain non-abelian o-groups or nilpotent i-groups;

(12) it is contained in the i-variety Wa of weakly-abelian i-groups;
(13) it is metabelian.
Recall that the group G = ZwrZ described earlier has the properties (2) and
(3). Now let I be the set of all infinite sequences a = (ao, al,"" an, .. .), where
an E {+1, -I}. For every sequence a E I, let us consider the linear order P a on
group G, defined uniquely by the following relations:

b ~ aa o ~ [a, W" ~ ... ~ [a, b, ... , b]a n ~ ... > e.

--.....--
n

For each n, let (an, Fa(n) be the quotient group G/'YnG with the naturally induced
order. It is not difficult to verify the following properties of (G, Pa):
(14) for sequences a = (ao, aI, ... ,) and -a =
(-ao, -al,' .. ,) the o-groups
(G, Pa) and (G, P- a ) are order isomorphic under an isomorphism cp such
= =
that cp(b) band cp(aa o ) a-a o ;
(15) the o-groups (an, Fa(n) and (an, F-a(n) are also order isomorphic;
(16) the o-group (G, Pa) is a subdirect product of the o-groups (an, Fa(n)
because n InG = e;
00

n=l
(17) if for some elements x, y E (G, Pa), I[x, yJl > e, then Ixl V Iyl = bScp
for some integer sEN and cp E ED E (abk) = fun«b),(a», the base
kEZ
subgroup of G.
We denote by a(n) the final segment (an, ... , an+l"") of the sequence a =
(ao, al,···, an+l, .. .).

Lemma 5.10 Let a = (ao, aI, ... , an-I, ... , an, ... ) be a sequence from I; let s, t E
Z be such that s > 0 and tan> 0; let cp, t/J E fun«b), (a» with t/J E rn+2G; and let

x = bScp, y = [a, b, ... , Wt/J.

--.....--
n

Then the subgroup gp(x, y) of the o-group (G, Pa) is order isomorphic to the o-group
(G, P a(n) or (G, p-a(n)'

Let us denote by j ~ I the set of all those sequences a such that ao = + 1. Now
consider the group H = fJaEi(G, Pa), and let G be the i-subgroup of H generated
by ii, b, where ii(a) = a and b(a) = b for all a.

Lemma 5.11 Ifforsome elements x,y E G l[x,yJl =I e, then C-gp(l[x,y]1, IxlVlyl)

(the i-subgroup of G generated by the elements I[x, yll and Ixl VIyl) is i-isomorphic
to G under an i-isomorphism which maps ii 1-+ I[x, y]1 and b 1-+ Ixl V Iyl.
12 Kopytov and Medvedev

Let H 1 = (Y IF be an ultrapower of the i-group O. The following statements

follow immediately from the definitions and Lemma 5.1l.
Lemma 5.12 Let u, v be elements of Hl = oJ IF and I[u, v]1 :F e. Then the i-
subgroup i-gp(l[u, v]l, lui V Iv I) is i-isomorphic to O.
Lemma 5.13 Let X be a non-abelian i-subgroup of the cartesian product TI,Er V,
where each V, is an ultrapower of the i-group G .Then for every pair of elements
u, v E X such that [u, v] :F e, l-gp(lul V Ivl, I[u, v]l) is i-isomorphic to G.
By Theorem 2.2, the i-quasivariety ql(O) generated by the l-group 0 coincides
with the class of l-groups SPPu(O), where S, P, and Pu are the closure opera-
tions of forming l-subgroups, cartesian products and ultraproducts. Therefore, if
X is a non-abelian l-group from ql(O), then by Lemma 5.13, X contains an i-
subgroup which is l-isomorphic to the i-group G. It follows that every non-abelian
i-quasi variety Ql ~ ql (0) contains the l-group 0 and so Ql = ql (0).
Theorem 5.14 (Medvedev [56](1992)) The i-quasivariety ql(G) has the following
properties:
(18) qt( G) covers the l-variety A of abelian i-groups in the lattice A of
i-quasivarieties;
(19) every a-group in ql(G) is abelian;
(20) every nilpotent l-group in ql(G) is abelian;
(21) ql(G) is contained in the i-variety Wa of weakly abelian l-groups.

The following series of covers of A was discovered by S. V. Varaksin [67](1993).

Let F = F(x, y) be the free group on x and y, and

F = 'Yl(F) ~ 'Y2(F) ~ ... ~ 'Yn(F) ~ 'Yn+l(F) ~ ...

be the lower central series of F. It is well known (see, e.g., Hall [28]) that each
element a of the factor-group 'Yn (F)I'Yn+l (F) can be uniquely represented in the
form a = 0:~1 ... o:~I'Yn+l (F), where 0:1, ... , 0:/ are basic commutators of weight n
and 0:1 -< 0:2 -< ... -< 0:/ in the standard order on the set of basic commutators of
the weight n.
Now we fix some infinite sequence <p = 101" ... ,10/, ... with each 10 = ±1, and
define a total lexicographic order Pn on the factor-group 'Yn(F)hn+l(F) by setting:

0:~''Yn+1(F) > 'Yn+l(F), .. ·,o:f''Yn+l(F) > 'Yn+l(F),

00

and O:~''Yn+l(F) « ... «0:;I'Yn+1(F). Since n 'Yn(F) = E, we can define a total

n=l
lexicographic order Q<.p onF by setting a > e, a E F if and only if a'Yno+l (F) >
'Yno+l (F) in the total order Pno , where no is the smallest integer n such that
a'Yn+l (F) :F 'Yn+l (F). It is clear that there exist 2N o different lexicographic total
orders Q<.p. Let I be the set of all sequences of ±l.
Quasivarieties and Varieties of Lattice-Ordered Groups 13

Theorem 5.15 (Varaksin [67](1993)) There exists a subset h ~ I of cardinality

2 N o and lexicograpically totally ordered free groups (F, Q ... ), (if' E II such that:
1) qe( F, Q ... ) covers A in the lattice A of f-quasivarieties;
2) all these covers ql(F, Q... ) are different and are contained in the f-variety Wa
of weakly-abelian f-groups

6 Covers in the lattice of R-quasivarieties

Here we will exhibit some f-quasivarieties which have no covers in the lattice A of
l-quasivarieties.
Let P be the set of all prime numbers and M ~ P. The f-quasivariety QM is
defined by the following system of implications ~M:

«Ixl V Iyl)-ll[x, y]I(lxl V Iyl) = I[x, yJIPk) :::;. ([x, y] = e),

where kEN, k 2:: 2,p E M ~ P. Let Gpk = gp(c, d I c-1dc = d pk }. Then the normal
closure D = gp(dU I g E Gpk} of the element d is isomorphic to the subgroup of the
rational numbers Q which consists of the elements %with a E Z and b = pkn, n E Z.
Thus, Gpk is the semi direct product of D and the infinite cyclic group (c). Since
D is isomorphic to a subgroup of the rational numbers Q, then D has only two
different total orders. Let P be a total order on D such that d> e. Now we define
a total order on Gpk by the rule: cid 2:: e iff i > 0 or i = 0 and d E D, d 2:: e in
o-group (D, P).

Theorem 6.1 (Medvedev [52](1985)) For each non-empty set M of prime numbers
there is no f-quasivariety covering the f-quasivariety QM in the lattice A of l-
quasivarieties.

Corollary 6.2 There are 2No different f-quasivarieties each of which has no covers
in the lattice A.

In the theory of quasivarieties (or varieties) of algebraic systems there is a well-

known connection between independent bases of implications (identities) of a qua-
sivariety (variety) and covers in the lattice of quasivarieties (varieties) of this quasi-
variety (variety). We formulate this connection for f-quasivarieties and f-varieties.

Proposition 6.3 (Gorbunov [14](1977)) Let V ~ W ~ £ be f-varieties (f-quasivar-

ieties) and let W be finitely based. Let V have an infinite independent basis of iden-
tities (implications). Then there is an infinite set of i-varieties (f-quasivarieties)
Ui, i = 1,2, ... , n, ... such that each Uj covers V in the lattice of f-varieties (f-
quasivarieties) and Uj ~ W, i = 1,2, ... , n, ....

Proof. Let 4> = {if'j Ij E N} be an infinite independent basis of identities

(implications) of V. Let
14 Kopytov and Medvedev

and let Vk be the £-variety (£-quasivariety) defined by the identities 'h, kEN.
n Vk 2 = V for all k 1 , k2 EN, kl '# k 2 . From the compactness
It is clear that Vk ,
theorem (see Burris and Sankappanavar [7] , page 212), it follows that there exists
a finite subset K ~ N such that Vk ~ W for all kEN \ K We claim that each
Vk (k E N \ K) contains a cover of V. Assume, on the contrary, that Vk has no
cover of V. Then there exists a descending chain of £-varieties (£-quasivarieties)
V ~ ... ~ .fa ~ ... ~ Vk ~ W for which W = n
X",. By our assumption, V is
erE!
defined in the £-variety (£-quasivariety) Vk by one identity (implication) 'Pk. By
the compactness theorem, the identity 'Pk follows from the identities defining the
£-variety (£-quasivariety) Xao' Thus, V = X ao , in contradiction to our assumption.

Corollary 6.4 Each £-quasivarzety QM has no independent basis of zmplications.

The proof follows immediately from Proposition 6.3.

7 f- Varieties

We begin by recalling a fundamental theorem due to G. Birkhoff (sec Burris and

Sankappanavar [7]).

Theorem 7.1 (Birkhoff [5](1935)) For a non-empty class IC of £-groups the follow-
ing statements are equivalent:
(a) IC is closed under the formation of Cartesian products, homomorphic
zmages and f-subgroups;
(b) K is equatwnally defined (that is, for some family of identttzes :E, IC IS
the class of all €-groups which satisfy all identities of :E).

For any class K of £-groups let H(K) denote the class of €-groups which are
homomorphic images of elements of IC. Let varlK be the £-variety generated by
K (that is, the smallest £-variety containing K). From Theorem 7.1 it follows that
varlIC = H 5P(K).
By a result of G. Birkhoff's [6](1942), the lattice of ideals of any £-group is
Brouwerian and so distributive. Therefore there is another description due to B.
J 6nsson, using ultraproducts, of the £-varicty generated by a class K.

Theorem 7.2 (.Jonsson [37](1967)) Lei V = varl(IC) be the £-vanety generated by

some class IC of £-groups. If A is a finitely subdirectly irreducible £-group in V, then
A E H5Pu (K) where II, 5, and P u are the operations of €-homomorphic images,
£-subgroups and uitraproducts, respective/yo Hence,

V = vari(K) = SP H 5Pu(K)'
Quasivarieties and Varieties of Lattice-Ordered Groups 15

We will denote the i-variety of all i-groups by £. and the i-variety of trivial
i-groups by E. Some i-varieties are defined by identities in the signature of group
theory alone, { . , -1 , e }. We call them the group i-varieties. For group i-varieties
we use the following notation:
(1) A is the i-variety of all abelian i-groups;
(2) n k is the i-variety of all nilpotent i-groups of class ~ k;
(3) Ak is the i-variety of all i-soluble i-groups of i-soluble
length ~ k;
As an example of an i-variety which is not a group i-variety, let X be the i-
= =
variety defined by the identities [x, y, z] e and [I[ x, y]11\ Iz I, t] e. It is easy to see
that G E X iff the i-ideal of G generated by the derived subgroup G' of G is central.
It is not difficult to construct an example of a totally ordered nilpotent group G of
class 2, such that the convex subgroup generated by G' is not central. Therefore,
X #n 2 . On the other hand, it is well known that every torsion-free nilpotent group
G has a total order such that the centre Z of G is a convex subgroup of G. Applying
this observation to a free nilpotent group of class 2, it is evident that every identity
of signature { . , -1 , e } which is valid in all i-groups from X is valid in n 2 , too.
Hence, the i-variety X cannot be defined by identities of signature { . , -1 , e }.
We next recall some very important i-varieties:
(4) N is the i-variety defined by the identity Ixl'lyll\ ly21'lx 2 1= Ixl'lyl;
(5) n is the i-variety defined by the identity (x 1\ y-lx-ly) Ve = e;
(6) Wa is the i-variety defined by the identity x- l lylxlyl-2 V e = e.

It is clear that the set L of all i-varieties is a complete lattice where for any
i-varieties X, Y the greatest lower bound, or meet, of X and Y is their set-theoretic
intersection and the least upper bound, or join, is the intersection of all i-varieties
containing both X and y.

8 The lattice of nilpotent i-varieties

For a long time there was the conjecture that the lattice of i-varieties of nilpo-
tent i-groups might be in some sense simpler than the lattice L of all i-varieties.
These hopes were dashed by the results of S. A. Gurchenkov [17, 19] and N. Va.
Medvedev [54].
Let us consider the following 2n (n E N) terms of the signature i;
Ul = (I[x, y]1 V tl) 1\ I[x, yW, WI = [Ixl V Iyl, UI],
U2 = (IWll V t2) 1\ IWlI2, W2 = [Ixl V Iyl, U2],

Un = (Iwn-ll V tn) 1\ IWn_11 2,

Now for every positive integer k ~ 3 we define the following identity
!k = wi 1\ ... 1\ wt 1\ [UI, U2]+ 1\ [U2, U3]+ 1\ ... 1\ [Uk-I, Uk]+ 1\ [Ul, Uk]+ = e.
16 Kopytov and Medvedev

Let Un be the i-variety defined by the identities ik = e, for k = 3,4, ... , n

and [x, y, z] = e. It is evident that

N. Ya. Medvedev proved that each containment is proper.

Therefore, we have the following result.

Theorem 8.1 (Medvedev [54](1989»

a) The i-variety U n Un of nilpotent i-groups of class::; 2 is not
nEN
finitely based.
b) The set of identities {tn = e 1 n E N} is independent.
c) The lattice of nilpotent i-varieties of class ::; 2 has uncountable
cardinality.

Corollary 8.2 The i-variety U = nnEN Un has infinite axiomatic rank.

Now let us consider the following terms of the signature i:

U1 I[x, y]1 /\ IZ11,

Uz (I[x, ydl V I[y, U1]1) /\ IZ21,
U3 (I[x, uz]1 V I[y, uzJl) /\ IZ31,
U4 I[x, u3]1 V I[y, u3]1,
v I[U1, xll- 1U3 /\ I[uz, yll-1 u4 /\ x /\ y,
11 V/\X 2y-1/\y 2 x- 1 .

For each pair of positive integers m, n let

Wm,n (v /\ x 2 y-1 /\ I[x, u3]1- 2n l[y, u3ll ffi

) Ve,
Um,n (v /\ x 2 y-1 /\ I[x, u3ll n l[y, u3ll- 2m ) Ve.

Now for each positive real number a > 0 let

L L
Ua A 2 /\ n /\ vardun,m = el n, mEN, n/m::; a},
L L
IF'' A2 /\ n /\ varl{ wn,m = el n, mEN, n/m::; a},
L L
B A2 /\ n/\ varduVe = e}.
Quasivarieties and Varieties of Lattice-Ordered Groups 17

Theorem 8.3 (Gurchenkov [20](1990))

a) The C-variety B has 2N o different covers in the lattice of nilpotent C-
varieties.
b) For each positive irrational real number 0:', neither of the C-varieties
Uo: and UO: has a finite basis of identities.
c) The lattice of nilpotent C-varieties is not Brouwerian.

It is known (see [Gurchenkov [19]) that the lattice L of all C-varieties has the
covering condition (i.e., each non-trivial C-variety has a cover in the lattice L) and
(see Medvedev [50]) the lattice Ln of all representable C-varieties does not have the
covering condition.
The following results were proved by S. A. Gurchenkov.

Theorem 8.4 (Gurchenkov [19](1990))

a) The lattice of all weakly abelian C-varieties does not have the covering
condition.
b) The lattice of all nilpotent C-varieties does not have the coverzng
condition.
c) For each k 2: 5, the lattice Lnk of nilpotent C-varieties of class :S k does
not have the covering condition.

By Proposition 6.3 each non-trivial C-variety without covers in the lattice of the
C-varieties contained in some finitely based i-variety has no independent basis of
identities. Therefore the following result is connected with the covers in the lattice
L nk •

Theorem 8.5 (Gurchenkov [18](1988),[19](1990)) There exists an C-variety X of

nilpotent i-groups of class $ 3 without an independent basis of identities.

Let us recall that an i-variety X is divisible if each C-group G E X can be

embedded in a divisible C-group G* EX. It is known that the C-variety £. of all
C-groups is divisible (Holland [30](1963)).

Theorem 8.6 (Gurchenkov [21](1991))

a) There are uncountably many nilpotent class 2 divisible i-varieties with
finite axiomatic rank.
b) There are uncountably many nilpotent class $ 3 C-varieties each of which
is not divisible.

In [23](1994), S. A. Gurchenkov constructed an infinite family of weakly abelian

i-varieties Mp, (p is a prime number) which are not generated by their nilpotent
i-groups.
18 Kopytov and Medvedev

Let us recall that a group G is called an Engel group if for each pair of elements
x, y E G there is an integer t = t(x, y) > 0 such that

[x,y, ... ,y)=e.

----- t

If the identity holds in a group G for some fixed positive integer t independent of x
and y, then G is called a i-Engel group.
For any pair x, y E G with x > e, let

The following results give necessary and sufficient conditions for representability
of Engel C-groups.

Theorem 8.7 (Medvedev [53)(1988)) Let G be an Engel C-group. Then G is rep-

resentable if and only if for every x, y E G, x > e, there exists a positive integer
t = t(x, y) such that the equality [u, y, ... , y) = e holds in G for all u E M(x, y).

-----
t

Theorem 8.8 (Gurchenkov [25)(1994)) Let G be an Engel C-group. Then varl(G) C

N if and only ifvarl(G) C n.
Corollary 8.9 Let G be a bounded t-Engel C-group. Then G is representable.

The following surprising result was obtained by Y. K. Kim and A. H. Rhemtulla

with the use of this Corollary.

Theorem 8.10 (Kim and Rhemtulla [41)(1992)) A bounded t-Engellattice-ordered

group is nilpotent.

9 Covers in the lattice of {-varieties

The smallest proper £-variety in the lattice L is the abelian C-variety A. Because
A is defined by a single equation, it follows from general principles that every
i-variety which properly contains A contains a cover of A in the lattice L of C-
varieties. All the covers of A containing non-representable C-groups are known;
they are the prime Scrimger varieties (see Reilly [63)). The covers of A consisting
of only representable £-groups may not be completely known. Medvedev [49](1977)
found three solvable covers of A which are representable (generated by o-groups),
and proved that they are the only solvable representable covers. Kopytov [43)(1985)
and Bergman [4](1984), independently, discovered two more representable covers.
Then Holland and Medvedev [33](1994) produced an uncountable collection of
representable covers of A. Each of these £-varieties is generated by a free group,
 Quasivarieties and Varieties of Lattice-Ordered Groups 19

totally ordered so that conjugation by certain elements produces large shifts of the
archimedean classes, either up or down, according to the pattern of any prescribed
infinite sequence of ±l's. Specifically, if a and b are free generators and e ~ a ~ b,
then there is generated a sequence of ± l's (corresponding to "up" and "down",
respectively) by considering the effect of conjugation first of b on a, then by the
larger of a and a b on the smaller, etc. The Kopytov-Bergman groups correspond to
the constant sequences of all + l's and all -1 's, respectively. Holland and Medvedev
were able to order the free group so that any given sequence of ±1 's can be produced.
Furthermore, the action of conjugating up or down can be captured by equations.
From this, the following theorem was proved.

Theorem 9.1 (W. C. Holland and N. Ya. Medvedev [33](1994)) There are an
uncountable number of representable covers of the abelian i-variety A in the lattice
of i-varieties.

Indeed, using a combinatorial argument of G. Bergman (personal communica-

tion), the cardinality in Theorem 9.1 can be shown to be 2No .
Medvedev [50](1982) defined a variety containing all representable covers of A.
Sometime later, M. Anderson, M. Darnel and T. Feil [2] also found a variety con-
taining all representable covers of A.
Let C be the i-variety defined by the infinite system of identities

(([Ixl A Iyl, Ixll V e) A (Ixl A Iyl))n A ((Ixl A Iyl) V Ixl-1(lxl A lyl)lxl) =

= (([Ix I A Iyl, Ixl] V e) A (Ixl A Iyl))n (n EN).
Theorem 9.2 (Anderson, Darnel, and Feil [2](1991)) The i-variety C contains all
covers of A contained in n.
We recall that a formula<p is universal if it is ofthe form VXl, ... , VXn 1/;, where 1/;
is quantifer-free. A class K is universal class if for some family of universal formulas
E, K is the class of all i-groups which satisfy all formulas of E. The question about
covers of i-varieties in the lattice T of universal classes of i-groups was considered
by Medvedev [57].
Let U be the i-variety defined by the following system of identities El :

a) (Ixl V lyl)-l(l[x, y]i A Itl)(lxl V Iyl) A (I[x, yll A Itl)n =

::: (I[x, yll A Itl)n, (n E N, n ~ 3),
b) (x A y-1x-ly) Ve = e,
c) [I[x, y]1 A It I, I [Xl, yIlI A It 1 III = e.
20 Kopytov and Medvedev

By it we denote the i'-variety defined by the system of identities 't :

a) (Ixl V lyl)-l(l[x, y]IA Itl)(lxl V Iyl) A (I[x, y]IA Itl)3 = (I[x, y]IA It1)3,
b) (xAy-1xy)Ve=e,
c) [l[x, y]IA It I, l[xI, YIlIA Itld = e.
It is clear that U S;; it and U I- it.
Theorem 9.3 (Medvedev [57](1992)) In the lattice T of universal classes of i'-
groups there are no universal classes W such that
a) W covers the E.-variety U in the lattice T;
b) WS;;it.
Corollary 9.4 The i-variety U has no independent basis of universal formulas.

M. R. Darnel [11](1994) studied non-representable covers of the i'- variety R

of all representable i'-groups. In that paper, a variation of the method used to
construct the Scrimger i'-varieties is developed that is shown to produce every non-
representable cover of any representable i'-variety. All non-representable covers
of any weakly abelian E.-variety and any i'-metabelian representable E.-variety are
described.
It is shown that any non-representable E.-variety is a subvariety of a certain
qua..<;i-representable E.-variety defined hy N. Reilly [62](1981).

10 The semigroup of i-varieties

It is known (Reilly [63](1989)) that the lattice-ordered semigroup L of of i'-varieties
satisfies the identity

N. Ya. Medvedev [55](1989) proved that the identity

U(/\ Vi) = /\UVi

JET JET

is not valid in the semi group L for an infinite index set I.

Let us recall that for the E.-variety R of representable E.- groups and for repre-
sentable E.-varieties U, V we set U * V = U . V n R.
N. Ya. Medvedev [54](1989) also proved that the identity

U * (/\Vi ) = /\U * Vi
JEI JEI
Quasivarieties and Varieties of Lattice-Ordered Groups 21

is not valid in the semi group Ln of representable i- varieties for an infinite index
set I.
As usual, let

AW = V A n,A*l = A, A*n+l = A*n *A for n E N

nEN

and

Li Si Ze [47](1990) proved the following statements;

(1)
00

R. (2)

The powers of the i-variety R of all representable i-groups were considered by

M. Darnel [11](1994). He proved the following statements:
a) Let V be any i-variety such that V C N and V f:. N. Then V ~ Rn for
some positive integer n.
b) An i-variety V is solvable if and only if V n R is solvable; and V is
i-solvable if and only if V n R is i-solvable.
For some finitely based varieties of i-groups U and V it is possible to prove the
existence of a finite basis of identities of the product UV. For example, we have the
following.

Theorem 10.1 (Litvinova [46](1994))Let n

k be the i-variety of all nilpotent i-
groups of class:::; k and C n be the i-variety defined by the identity

[x,y]=e.

Then the i-varieties nk . V and (C n n A2) . V are finitely based for each finitely
based i-variety V.

11 Ordered permutation groups and i-varieties

Let S1 be a totally ordered set and A(S1) be the i-group of all order preserving
permutations of S1. One can ask which i- varieties can be generated by the i-groups
A(S1). Such i-varieties are now completely known. This remarkable result is due to
W. C. Holland.
22 Kopytov and Medvedev

Theorem 11.1 ( Holland [32](1985)) Let A(Q) be the ji-group of all order preserv-
ing permutations of the totally order'ed set Q. Then vare(A(Q)) IS eliher the ji-variety
£ of all ji- groups, the ji-variety N of all normal-valued ji-groups, the ji-variety £ of
trivial f!-groups, or An for some positive integer n.

Let G be a right-ordered group and R : G ---+ A(G,~) be the right regular

representation of the right-ordered group (G, ~). Then R(G) is a partially ordered
subgroup of the i-group A( G, ~) of all order preserving permutations of the totally
ordered set (G, ~). Let G* be the f!-subgroup of A(G,~) generated by the subgroup
R(G). Some properties of G* are inherited from G. For example, if a right-ordered
group Gis Conradian, thcn G* is a normal-valued i-group (Kopytov [42](1979)).
S. V. Varaksin [65], M, Darnel and A. M. W. Glass [13], and M. Darnel [8] inves-
tigated the relation between thc group identities of G and the ji-group identities of
G*
Theorem 11.2 (Varaksin [65](1989)) f,et (G,~) be a right-ordered group and sup-
pose there is a finite subnormal series

of convex subgroups, all factors Gk/Gk-l being a-groups (k = 1, ... , n - 1).

Ifvare(Gk/Gk-J) =
Xk, then var/G' <;; Xl' X 2 ·· ,Xn'

Corollary 11.3 (Darnel [8](1991)) Let G be a fimtely generated torsion-free nilpo-

tent group of Hirsch length n. Then the free lattice-ordered group F( G) over G is
i-solvable of rank at most n.

Corollary 11.4 (Darnel and Glass [13](1989)) The free lallice-ordered group over
any m-generator torsion-free nilpotent group of class 2 is i-solvable of length at

most ( r; ) + 1.
The varieties of lattice-ordered groups generated by simple i-groups were consi-
dered hy S. V. Varaksin.

Theorem 11.5 (Varaksin [66J(1990)) Let G be an C-slmple non-representable ji-

group. Then the f!-variety generated by G is /'l or £.

Corollary 11.6 EVeT'y (group-)solvable f!-simple i-group is abelian.

Let us recall that the i-variety X has the amalgamation property if for all f!-
groupsA, B, C E X and ernbeddings 0' : A --+ B, /1: A --+ C, there exists an i-
group D in X and embeddings 'P : B --+ D, 'l/!: C --+ D such that 'PO' = 'l/!/1.
The quintuple (A, B, C, 0', /1) is called a V-formation in X and the triple (0', /1, 0:)
is an amalgamation in X of this formation. An algebra A E X is said to be an
Quasivarieties and Varieties of Lattice-Ordered Groups 23

amalgamation base for X if every V-formation (A, B, C, (J', J.L) in X has an amalga-
mation in X. The amalgamation class of X, AmalX, is the class consisting of all
amalgamation bases for X.
S. A. Gurchenkov investigated i-varieties with the amalgamation property. He
proved the following remarkable result.

Theorem 11.7 (Gurchenkov [24](1994))

a) Each non-representable i-variety fails the amalgamation property.
b) Each i-variety X such that X J A2 n Wa fails the amalgamation
property.
c) If the totally ordered additive group Z of integers is contained in AmalX
and X n n # X I then X contains the i-variety N of all normal-valued
i-groups.

An R-variety V has the lex-property if V = vart( {G 'X ZIG E V}). M. E. Hansen

in [29](1991) studied i-varieties with the lex-property. She proved that each quasi-
representable R-variety defined by N. Reilly in [62](1981) has the lex-property. On
the other hand, the Feil R-varieties and the Medvedev i-varieties (for both of these,
see Reily [63]) do not have it.
Let A and B be totally ordered groups. The restricted wreath product AwrB
can be totally ordered in two ways. Let 9 E AwrB, 9 = (j, b), where j: B ~ A
has finite support. Define (j, b) > (e, e) if b > e or both b = e and j(b o ) > e where
bo = max(supp(j)). This gives an o-group denoted by A ;-r B. The o-group A V/r
B is defined analogously, except (j, b) > (e, e) if b > e or both b = e and j(bd > e
where bl = min(suppj). One solvable Medvedev cover of A, denoted by M+, is
generated by Z ;-r z, where Z is the group of integers with the usual order, and
the other, denoted by M- is generated by Z V/r Z. Let vi = vart«Z V/r Z) 'X Z
and Vt- = vart«Z ;-r Z) 'X Z. The following interesting result was obtained by M.
Huss and later by Darnel [10]( 1993).

Theorem 11.8 (Huss [35](1984)) vi = Vt-.

References
[1] M. Anderson and T. Feil, Lattice-Ordered Groups: An Introduction, Kluwer
Academic Publ., Dordrecht, 1988.
[2] M. Anderson, M. R. Darnel, and T. Feil, A variety of lattice-ordered groups
containing all representable covers of the abelian variety, Order 7( 1991), 401-
405.

[3] A. K. Arora, Quasi-varieties of lattice-ordered groups, Algebra Universalis

20(1985),34-50.
24 Kopytov and Medvedev

[4] G. M. Bergman, Specially ordered groups, Comm. Algebra 12(1984), 2315-

2333.
[5] G. Birkhoff, On the structure of abstract algebras, Proc. Cambridge Phil. Soc.
31(1935), 433-454.
[6] G. Birkhoff Lattice-ordered groups, Ann. Math. 43(1942), 298-331.
[7] S. Burris and H. P. Sankappanavar, A Course in Universal Algebra, Graduate
Texts in Mathematics 78, Springer-Verlag, New York-Heidelberg-Berlin, 1981.
[8] M. R. Darnel, The free lattice-ordered group over a nilpotent group, Proc. Amer-
ican Math. Soc. 111 (1991), 301-307.

[9] M. R. Darnel, Disjoint conjugate chains, in Ordered Algebraic Structures (The

1991 Conrad conference), ed. J. Martinez and C. Holland, Kluwer Academic
Pub., Dordrecht, The Netherlands, (1992), 31-49.
[10] M. R. Darnel, Varieties minimal over representable varieties of lattice-ordered
groups, Comm. Algebra 21(1993), 2637-2665.

[11] M. R. Darnel, Cyclic extension of the Medvedev ordered groups, Czechoslovak

Math. J. 43(118)(1993),193-204.
[12] M. R. Darnel, Powers of the representable variety of lattice-ordered groups,
Algebra Universalis, (to appear).
[13J M. R. Darnel and A. M. W. Glass, Commutator relations and identities in
lattice-ordered groups, Michigan Math. J. 36(1989), 203-21l.

[14] V. A. Gorbunov, Covers in the lattice of quasivaneties and independent ax-

iomatizability, (in Russian), Algebra i Logika, 16(1977), 507-548.

[15] S. A. Gurchenkov, Varieties of f.-groups with the identity [xP, yPJ = e have finite
basis, Algebra and Logic 21(1984), 20-35.

[16] S. A. Gurchenkov, On covers in the lattice of f.-varieties, (in Russian), Mat.

Zametki, 35(1984), 677-684.
[17] S. A. Gurchenkov, The lattice of quasivarieties of 2-nilpotent f.-groups is not
distributive, (in Russian), Siberian school on varieties of algebraic systems,
Barnaul, (1988), 21-24.
[18] S. A. Gurchenkov, On the theory of varieties of lattice-ordered groups, (in Rus-
sian), Algebra i Logika 27(1988), 249-273.
[19] S. A. Gurchenkov, The lattice of varieties of weakly abelian lattice-ordered
groups does not have the covering condition, (in Russian), Mat. Zametki
47(1990),35-40.
Quasivarieties and Varieties of Lattice-Ordered Groups 25

[20] S. A. Gurchenkov, The lattice of nilpotent i-varieties is not Browerian and it

has the cardinality of the continuum, (in Russian), Izv. Visch. Uchebn. Zaved.
Math. 34(1990), 17-22.

[21] S. A. Gurchenkov, On three questions of the theory of i-varieties, (in Russian),

Czechoslovak Math. J. 41(1991),405-410.

[22] S. A. Gurchenkov, On completion of invariant locally nilpotent subgroups of

totally ordered groups, (in Russian), Mat. Zametki, 51(1992),35-39.

[23] S. A. Gurchenkov, About varieties of weakly abelian i-groups, to appear in

Math. Slovaca.

[24] S. A. Gurchenkov, On amalgamation in i-varieties, (in Russian), to appear in

Algebra i Logika.

[25] S. A. Gurchenkov, On Engel lattice-ordered groups, (in Russian), to appear in

Algebra i Logika.

[26] S. A. Gurchenkov and V. M. Kopytov, Description of covers of the variety of

Abelian lattice-ordered groups, Sibirskii Mat. Zh. 28(1987), 66-69.

[27] Yu. S. Gurevich and A. I. Kokorin, Universal equivalence of ordered abelian

groups, (in Russian), Algebra i Logika 2(1963),37-39.

[28] M. Hall, The Theory of Groups, Macmillan, New York, 1959.

[29] M. E. Hansen (M. Huss), The lex property of varieties of lattice-ordered groups,
Algebra Universalis 28(1991),535-548.

[30] W. C. Holland, The lattice-ordered group of automorphisms of an ordered set,

Michigan Math. J. 10(1963),399-408.

[31] W. C. Holland, The largest proper variety of lattice ordered groups, Proc. Amer-
ican Math. Soc. 57(1976), 25-28.

[32] W. C. Holland, Varieties of automorphism groups of orders, Trans. American

Math. Soc. 288(1985),755-763.

[33] W. C. Holland and N. Ya. Medvedev, A very large class of small varieties of
lattice-ordered groups, Comm. Algebra 22(2)(1994), 551-578.

[34] W. C. Holland, A. H. Mekler and N. R. Reilly, Varieties of lattice-ordered groups

in which prime powers commute, Algebra Universalis 23(1986), 196-214.

[35] M. Huss (M. E. Hansen), Varieties of Lattice-Ordered Groups, Ph. D. Disser-

tation, Simon Fraser University, (1984).
26 Kopytov and Medvedev

[36] O. V. Isaeva and N. Ya. Medvedev, Covers in the lattice of quasivarieties of

i-groups, (in Russian), Sibirsk. Mat. Zh. 33(1992), 102-107.

[37] B. Jonsson, Algebras whose congruence lattices are distributive, Math. Scand.
21(1967),110-121.

[38] M.1. Kargapolov and Yu. 1. Merzljakov, Fundamentals of the Theory of Groups,
Springer-Verlag, Berlin, 1979.

[39] N. G. Khisamiev, Universal theory of lattice-ordered abelian groups, (in Rus-

sian), Algebra i Logika 5(1966),71-76.

[40] N. G. Khisamiev and A. I. Kokorin, An elementary classification of lattice-

ordered abelian groups with a finite number of fibers, (in Russian), Algebra i
Logika 5(1966), 41-50.

[41] Y. K. Kim and A. H. Rhemtulla, Orderable groups satisfying an Engel con-

dition, in Ordered Algebraic Structures (The 1991 Conrad conference), ed. J.
Martinez and C. Holland, Kluwer Academic Pub., Dordrecht, The Netherlands,
(1992),31-49.

[42] V. M. Kopytov, Free lattice-ordered groups, Algebra and Logic 18(1979), 259-
270.

[43] V. M. Kopytov, A non-abelian variety of lattice-ordered groups in which every

solvable i-group is abelian, Mat. Sbornik 126/168(1985),247-266.

[44] V. M. Kopytov and N. Ya. Medvedev, The structure of varieties of lattice-

ordered groups, in Algebra and Order. Proc. First Int. Symp. Ordered Alge-
braic Structures, Luminy-Marseilles, ed. S. Wolfenstein, R&E 14, Helderman,
Berlin, (1984), 35-46.

[45] V. M Kopytov and N. Ya. Medvedev, Varieties of lattice-ordered groups, (in

Russian), Uspechi Mat. Nauk 40: 6(256)(1985),117-128; translated into En-
glish in Russian Mathematical Surveys 40: 6(256)(1985),97-110.

[46] M. B. Litvinova, On the product of finitely based varieties of lattice-ordered

groups, (in Russian), to appear in Algebra i Logika.

[47] Li Si Ze, On the varieties of representable i-groups, (in Chinese with English
summary), J. Math. (Wuhan), 10(1990),321-324.

[48] J. Martinez, Varieties of lattice-ordered groups, Math. Zeit. 137(1974), 265-

284.

[49] N. Ya. Medvedev, The lattices of varieties of lattice-ordered groups and Lie
algebras, Algebra and Logic 16(1977), 27-31.
Quasivarieties and Varieties of Lattice-Ordered Groups 27

[50) N. Ya. Medvedev, On the theory of varieties of lattice ordered groups, (in Rus-
sian), Czechoslovak Math. J. 32(1982),364-372.

[51) N. Ya. Medvedev, Free products of i-groups, (in Russian), Algebra i Logika
23(1984),493-511.

[52) N. Ya. Medvedev, Quasivarieties of i-groups and groups, (in Russian), Sibirsk.
Mat. Zh. 26(1985), 111-117.

[53) N. Ya. Medvedev, On o-approximability of bounded Engel i-groups, (in Rus-

sian), Algebra i Logika 27(1988), 418-421.

[54) N. Ya. Medvedev, On nilpotent lattice-ordered groups, (in Russian), Mat. Za-
metki 45(1989), 72-79.

[55) N. Ya. Medvedev, On infinite distributivity in the lattice of i-varieties, (in

Russian), Sibirsk. Mat. Zh. 30(1989), 216-220.

[56) N. Ya. Medvedev, On covers in the lattice of quasivarieties of i-groups, in

Ordered Algebraic Structures (The 1991 Conrad Conference), ed. J. Martinez
and C. Holland, Kluwer Academic Pub., Dordrecht, The Netherlands, (1992),
81-98.

[57) N. Ya. Medvedev, Independent axiomatization of varieties of lattice-ordered

groups, Czechoslovak Math. J. 42(1992), 53-57.

[58) N. Ya. Medvedev, HSPf.SHPS for metabelian representable i-groups, Algebra

Universalis 31(1994), 151-156.

[59) R. T. Botto-Mura and A. H. Rhemtulla, Notes on Orderable Groups, University

of Alberta, Edmonton, 1975; published as Orderable Groups, Lecture Notes in
Pure and Applied Math. 27, Marcel Dekker, New York, 1977.

[60] W. B. Powell and C. Tsinakis, The failure of the amalgamation property for
representable varieties of i-groups, Proc. Cambridge Phil. Soc. 106(1989),439-
444.

[61] W. B. Powell and C. Tsinakis, Covers of the variety of abelian i-groups, Comm.
Algebra 17(1989), 2461-2468.

[62] N. R. Reilly, A subsemilattice of the lattice of varieties of lattice-ordered groups,

Canadian J. Math. 33(1981), 1309-1318.

[63] N. R. Reilly, Varieties of lattice-ordered groups, in Lattice-Ordered Groups;

Advances and Techniques, ed. A.M.W. Glass and W. C. Holland, Kluwer Aca-
demic Publishers, Dordrecht/Boston/London, 1989, 228-277.
28 Kopytov and Medvedev

[64] D. A. Van Rie, Quasivarieties of i-metabelian lattice-ordered groups, Doctoral

Thesis, Bowling Green State University, 1991.
[65] S. V. Varaksin, Lattice-ordered groups constructed from right-ordered groups,
(in Russian), Algebra i Logika 28(1989), 524-533.
[66] S. V. Varaksin, Varieties generated by simple i-groups, (in Russian), Sibirsk.
Mat. Zh. 31(1990),167-180.

[67] S. V. Varaksin, A continuum of minimal quasivarieties of i-groups, (in Rus-

sian), Sibirsk. Mat. Zh. 34(1993) N4, 41-49.
 Lattice-ordered Permutation Groups:
The Structure Theory
Stephen H. McCleary

Bowling Green State University

Bowling Green, Ohio 43403, USA
Email: [email protected]

1 Introduction
This survey of lattice-ordered permutation groups focuses especially on their struc-
ture theory and on their relation to unordered infinite permutation groups. How-
ever, the survey is essentially self-contained. The reader will need only a little
familiarity with unordered permutation groups, and none at all with lattice-ordered
groups.
Let Q be a chain (totally ordered set). The group A(Q) of automorphisms (order-
preserving permutations) of Q acquires from Q the pointwise order, i.e., f ~ g iff
of ~ ag for all a E Q. It is easily checked that this order makes A(Q) a lattice-
ordered group (f-group), i.e., it makes A(Q) a lattice and the order is preserved by
multiplication on both sides (f ~ g implies hf ~ hg and fh ::; gh). The lattice
operations are given by a(f V g) = max{ af, ag} and dually, and thus are also
pointwise; see Figure l.
A subgroup G of A(Q) which is also a sublattice (closed under the lattice op-
erations) is called an f-subgroup, and (G, Q) is called a lattice-ordered permutation
group (f-permutation group). Some parts of the structure theory apply more gener-
ally to ordered permutation groups (with the order inherited from A(Q) but without
the sublattice requirement).
For any Q, the elements of A(Q) are homeomorphisms with respect to the order
topology on Q. Especially important is the case in which Q is the real line R.
Here every homeomorphism either preserves or reverses order, and of course A(R)
consists of those homeomorphisms doing the former. The study of A(R) goes back
at least to 1924, when H. Kneser [23] proved that any two positive elements without
fixed points must be conjugate. Later, f-permutation groups served as examples of
f-groups.
But the main impetus to the study of f-permutation groups has been Charles

29
w. C. Holland (ed.), Ordered Groups and Infinite Permutation Groups, 29-62.
© 1996 Kluwer Academic Publishers.
30 Stephen H. McCleary

J
9
JVg

Figure 1: J V9

Holland's analogue of Cayley's Theorem. The Holland Representation Theorem

[14] says that every €-group is (isomorphic as an €-group to) an €-subgroup G of
some A(O). This made obvious some elementary facts about €-groups which were
already known, e.g., that they are torsion free and that the underlying lattice is
distributive. The distributiv-ity is acquired from that of the chain 0 since
a((J V g) 1\ h) min{max{aJ, ag}, ah}
max{rrlin{aJ,ah},min{ag,ah}}
a((J 1\ h) V (g 1\ h)).
Furthermore, the Representation Theorem led Holland to deeper discoveries, for
example that every €-group can be embedded in a divisible €-group (extra care
being required in the choice of 0 in order to make A(O) divisible).
However, this survey deals directly with €-permutation groups (G, 0) rather than
with abstract €-groups. It includes a detailed description (with proofs) of the struc-
ture theory of transitive €-permutation groups, and several important applications.
The power of this theory derives from the interplay between its two complementary
parts:
(1) The collection of convex congruences (i.e., congruences having convex
classes) of a transitive €-permutation group (G, 0) is totally ordered by
refinement. In some arguments this affords a sort of "reduction" to the
"a-primitive" case. ("a-primitive" is the ordered analogue of "primitive",
and means that there are no proper convex congruences.).
(2) The classification of the a-primitive (G, O)'s, which makes the reduc-
tion useful.
Lattice-ordered permutation groups 31

The survey also offers a brief sketch of the more general intransitive structure
theory, with precise statements of just enough of it to enable one to see that many
applications of the general theory proceed almost exactly as in the transitive case.
This is done in such a way that a reader wishing to consider just the transitive case
can easily do so.
Following this is a section indicating which results of the structure theory (e.g.,
(1)) hold for ordered permutation groups. Even more of the results hold for coherent
i-permutation groups (G, r2), meaning that whenever ex < (3 and exg = (3 for some
9 E G then exh = (3 for some h E G+. (Here e denotes the identity and G+ = {g E
G I e <::: g} is the set of positive elements of G.) A word of caution: An ordered
permutation group (G, r2) for which the pointwise order makes G a lattice can fail
to be an i-permutation group because the lattice operations can differ from the
(pointwise) lattice operations of A(r2).
R, Q, and Z will denote the chains of real numbers, rational numbers, and
integers, resp., and N the set of natural numbers. The chain r2 will be assumed to
contain more than one element.
The standard reference in i-permutation groups is the book [8] by Andrew Glass,
and it contains most of the material presented here, along with an extensive an-
notated bibliography. Also, a shorter survey of the area by Holland appears as a
chapter in [9].

2 Examples
Here are some examples of transitive i-permutation groups. Transitivity prevents
r2 from having a largest or a smallest point, and in particular forces r2 to be infinite.
It also forces r2 to be either dense in itself or discrete. The first three examples will
turn out to illustrate the three types of o-primitivity, and the fourth will illustrate
a pathological subtype of the first type. The orbits of the stabilizer subgroup Go: =
{g E G I exg = ex} of a point ex E r2 will play an important role, and we precede
the examples with a brief discussion of these orbits. A subset E of r2 is convex if
0"1 <::: W <::: 0"2 (O"i E E, wE r2) implies wEE.

Proposition 201 Let (G, r2) be a transitive i-permutation group, and let ex E r2.
The orbits of Go: are convex.

Proof. Suppose 0"1 <::: w <::: 0"2 and that 0"1g = 0"2 for some 9 E G n . By transitivity,
0"1 h = w for some h E G. Then 9 /\ (h V e) moves 0"1 to wand fixes ex. See Figure 2.

Because of the previous proposition, the set of orbits of Go: inherits a total order
from r2 in the obvious fashion. Any non-singleton orbit of Ge; must be infinite, and
indeed can contain no largest or smallest point; such orbits are called long orbits.
32 Stephen H. McCleary

9
9
Q
Figure 2: Proof of the convexity of the orbits of Get

By transitivity this chain, together with the distinction between long orbits and
fixed points, is independent (up to isomorphism) of the choice of a. (This provides
a chance to emphasize our reliance on the trivial fact that when 9 : a1k f--t a2k then
k- 1gk : 0'1 I-t 0'2; and of course we have k- 1G et k = G etk . We note also that inner
(group) automorphisrns of i-groups are automatically i-group automorphisrns.)
Although we shall not use the information, we mention that the chain of positive
orbits of Get (those to the right of {a}) is anti-order-isomorphic to the chain of
negative orbits via the map b. -+ b.' from b. to its paired orbit b.' = {ag I a E b.g}.
Warning: It is possible for a long orbit of GO'. to be paired with a fixed point.

Exam.ple 1 (G, 0) = (A(R), R) [o-2-transitive]

This (G, 0) is 0-2-tmnsitive, i.e., for all al < a2 and {31 < f32 in 0, there exists
9 E G such that alg = (31 and a2g = (32 (and 101 > 2). For any 0-2-transitive
(G, 0), there are clearly only three orbits of GO'., namely {a} itself, a positive long
orbit consisting of all points to the right of a, and a similar negative long orbit.
We define o-n-transitivity (n E N) similarly. In contrast to unordered permu-
tation groups we have

Proposition 202 An 0-2-tmnsitive f-permutation group is o-n-tmnsitive for all n.

Proof. Let CG, 0) be 0-2-transitive. Let al < a2 < ... < an and (31 < (32 < ... <
(3n in o. By induction there exists 9 E G such that aig = {3i, i = 1, ... , n - 1. We
suppose ang 2: (3n, the other case being similar. By 0-2-transitivity, there exists
hE G such that alh = (3n-l (forcing ai 2: (3i, i = 1, ... , n -1) and anh = (3n. Then
O:i(g 1\ h) = (3i, i = 1, ... , n.

Exam.ple 2 (G,O) = the right regular representation of a subgroup of the additive

reals CR, ::;, +) [regular]

Here the orbits of G a are all singletons. If we take the subgroup to be Z, then
each automorphism of the ordered set Z is simply translation by some integer, so
that G = ACO).
In Example 2, we could have been more general, citing the right regular represen-
tation of any totally ordered group (o-group). However, the example as stated does
include all Archimedean o-groups since Holder's Theorem (1901) tells us that all
Lattice-ordered permutation groups 33

Archimedean o-groups are (up to isomorphism) o-subgroups of the additive reals.

(G is Archimedean if for all 0 < a < bEG there exists n E N such that na 2:: b.)

Example 3 G = {g E A(R) I (a + l}g = ag + 1 'Va E R} [periodic]

These are the elements of "period" one (though in a slightly non-standard sense-
see Figure 3). Any 9 E G maps [0,1) onto some other interval of length one, and
by periodicity that action uniquely determines 9 throughout R. The fixed points
of Go are the integers, and the long orbits are the intervals (n, n + 1), n E Z. As we
shall prove later, all o-primitive transitive i-permutation groups which are neither
0-2-transitive nor regular look very much like this one.

...
R

..
Figure 3: Periodicity

Example 4 G = {g E A(R) 13ng EN such that (a + ng)g = o:.g + ng 'Va E R},

with the period ng depending on 9 [pathologically o-2-transitive]

Like Example 1, (G, R) is 0-2-transitive. However, it is pathologically 0-2-

tmnsitive, meaning that it has no element =I=- e of bOWlded support. (The sup-
port of 9 means {a E R Iag =I=- a}, and is denoted by supp(g).) Pathologically
0-2-transitive (G, O)'s cause a considerable amoWlt of trouble.
VVhen (G, 0) is non-pathologically 0-2-transitive, every non-singleton interval .6.
of 0 supports some e =I=- f E G+. For given e =I=- 9 E G, let h = 9 V g-l E G+\{e}.
Picking a ~ supp(h) ~ T and picking kEG such that ak, Tk E .6., we have
supp(k- 1 hk) = (supp(h))k S;; .6..
34 Stephen H. McCleary

Exrunple 5 0 z-x--R, the lexicographic product ordered from the right; G

A(O).

The set of local copies of Z which replace the points of R in the formation of 0
is "respected" by G. That is, for each 9 E G the image 6g of one of these copies
is another one of the copies (since two points of 0 lie in the same copy iff there
are only finitely many points of 0 between them). Thus each 9 E G induces in
the obvious fashion an order-preserving permutation g of the chain of copies of Z.
In effect g E A(R), and of course locally 9 induces an integer translation t(3 for
each f3 E R. For 0: = (0,0), the positive fixed points of Gn consist of the positive
integers in the oth copy of Z , and the points to the right of the oth copy form the
one other positive orbit.

3 The Holland Representation Theorem

Now we revisit abstract £-groups and the Holland Representation Theorem [14]. A
subgroup H of an £-group G is convex if hI :::; 9 :::; h2 (hi E H) implies 9 E H; and
is prime if gl 1\ g2 E H implies gl E H or g2 E H.
In an £-permutation group, every Ga is a prime convex £-subgroup of G. More-
over, the orbit o:G is in one-to-one correspondence with the set R( G n ) of right cosets
Gng of G n via o:g f-+ Gag. This becomes an isomorphism of chains if we define

(If o:gl :::; o:g2, take k1 = g1 and k2 = gl V g2.)

An £-homomorphism from G 1 into G2 is a group homomorphism which also
preserves the lattice operations The subgroups of G which occur as kernels of £-
homomorphisms are precisely the normal convex £-subgroups, known as €-ideals.

Theorem 301 (Holland Representation Theorem) Every €-group G is an €-sub-

group of A(O) for some chain O.

In outline, Holland's proof goes like this. For each e cI 9 E G, pick a convex
£-subgroup Pg maximal with respect to not containing g; because of the maximality,
P q is prime. The crux of the argument is that primeness guarantees that setting

gives a total order on R(Pg), just as for G a above. Let Og be the chain R(Pg),
and let G act on Og by right translation. This gives an £-homomorphism cPg : G -+
A(Og). The stabilizer subgroup of Pg E Og is Pg itself, and ker(cpg) = core(Pg), the
intersection of all conjugates of Pg. In general, no one cPg will be faithful. However,
Lattice-ordered permutation groups 35

pasting together (disjoint copies of) the various ng'S gives a master chain n in which
the ng'S are convex and on which G acts faithfully.

The proof gives some extra information: There is a representation of G which

can be obtained by pasting together various transitive actions, i.e., for which the
orbits (the ny's above) are convex.
The proof also tells us which subgroups of an i-group G can occur as stabilizer
subgroups in representations of G as i-permutation groups, namely the prime convex
l-subgroups. (Even if P is not one of the Py's, R(F) can be adjoined to the end
of n.) We also discover that the P's which can occur as stabilizer subgroups in
representations of G as transitive l-permutation groups are those having trivial
core (since clearly when (G, n) is transitive, coree Go,} = nkEG Gak = {e}). Such
P's are known as representing subgroups, and of course the G's having transitive
representations are precisely those having representing subgroups.
We mention here an easy identity
9 = (g V e)(g 1\ e) = (g V e)(g-l V e)-I,

which is easily verified for i-permutation groups by computing O'.g, splitting into
the cases O'.g 2': a and O'.g < O'.. Then by the Holland Representation Theorem this
identity holds also for abstract i-groups. In particular, every l-group is generated
as a group by G+.

4 Doubly transitive A(D)

Let n be an arbitrary chain, let f E A(n), and let 8 E supp(f). The convexification
Conv{8r In E Z}, i.e.,

{w E n 18f-n ~ w ~ 8r for some n E Z},

is called a supporting interval of f (positive if 8f > 8, negative if 8f < 8). Since
{8 fTn I m E Z} is coterminal in this supporting interval, the interval and its parity
are independent of which of its points 8 is used to define it. The element f in Figure
1 has one supporting interval of each parity. The set of supporting intervals of any
f has a natural total order, namely .6. 1 < .6. 2 iff 81 < 82 for all 81 E .6.1 , 82 E .6.2.
An element b having exactly one supporting interval is called a bump or a convex
cycle. Each f E A(n)+ is uniquely the supremum of a (possibly infinite) pairwise
disjoint set of positive bumps, where gl, g2 E A(n)+ are called disjoint if their
supports are disjoint, or equivalently, if gl 1\ g2 = e.
Any conjugate of a bump is a bump of the same parity. This is a special case
of something more general: The landscape £(f) of any f E A(n) means the chain
of supporting intervals and fixed points of f, with positive supporting intervals,
36 Stephen H. McCleary

negative supporting intervals, and fixed points distinguished; and conjugation by

any h E A(O) induces an isomorphism from £(f) onto £(h- 1 fh). The following
conjugacy lemma provides a converse when A(O) is o-2-transitive. The lemma is
due to Holland [14], special cases having been proved previously by Kneser [23] and
by J. Schreier and S. Ulam [40j. The lemma tells us, for example, that in A(Q)
there are exactly nine conjugacy classes of positive bumps (the inf of supp(f) can
be rational, irrational, or -00, and dually).

Lemma 401 Suppose A(O) is o-2-transitive. Then f and g are conjugate in A(O)
if and only if £(f) and £(g) are isomorphic.

Proof Let 'ljJ be a landscape isomorphism from £(f) onto Leg). Let D..f be any
supporting interval of f. Assume D..f is positive, the negative case being similar. Let
D.. g = D..f'ljJ, which is also positive. Pick a E D..f, (3 E D.. g • By o-2-transitivity we can
pick an order-isomorphism ¢Yo from the interval [a, oJ) onto [(3, (3g); see Figure 4.

6.f
A

¢Yo ¢Yl
A A
( lr \
a af aj2

l
y

D.. g
Figure 4: The definition of the conjugator cP

We extend cPo to 6. f in the only way which can possibly yield the desired conjugator:
We define cPn+l on [ar,ar+l) to be cPn+l = f-1cPng (n = 1,2, ... ), and similarly
in the negative direction. Form cPD., : 6.f ---t D.. g by patching together the various
¢Yn's. Now form cP E A(O) by patching together the cPD., 's for the various supporting
intervals D..f of f and sending each fixed point w of f to w'ljJ. Then cP E A(O) and
¢y- 1 f¢Y = g.
Lattice-ordered permutation groups 37

Note that if f and 9 both have support bounded above (below), the same can
be arranged also for the conjugator </J.
Let L(O) consist of those elements of A(O) having supports bounded above
("living on the left"). Let R(O) be the dual. Let B(O) = L(O) n R(O), the
elements of bounded support. Always these are £-ideals of A(O).
Various parts of the following theorem were contributed by G. Higman [13], J.
T. Lloyd [25], and Holland [14].

Theorem 4.2 Let A(O) be 0-2-transitive. Then every normal subgroup of A(O)
is an £-ideal. Moreover, B(O) is a simple group and is the smallest proper normal
subgroup of A(O). If 0 has a countable coterminal subset (bounded neither above
nor below), then B(O), L(O), and R(O) are the only proper normal subgroups of
A(O), and B(O) is the only proper normal subgroup of either L(O) or R(O).

Proof that B(O) is simple. Let e =I b E B(O), and let 9 E B(O). Pick a
supporting interval ~ of b. Pick 0" < supp(g) < T, and use 0-2-transitivity to pick
k E A(O) such that O"k, Tk E ~; and arrange in addition that k E B(O) by using 0-
4-transitivity (see Proposition 2.2) to make k fix some 0"1 < 0" and some Tl > T and
then changing k to act like the identity outside [0"1, Tl]. Then ~ is also a supporting
interval of b(k- 1 gk), and band b(k- 1 gk) have the same landscape. By Lemma 4.1
ff., band b(k- 1 gk) are conjugate in B(O). It follows that 9 is the product of a
conjugate of b- 1 by a conjugate of b. Therefore B(O) is simple.

When the countable coterminality hypothesis in Theorem 4.2 is not satisfied,

there are necessarily many other normal subgroups of A(O)---see R. N. Ball [1],
Ball and M. Droste [2], and Droste and S. Shelah [6].
Another consequence of Lemma 4.1 is

Corollary 4.3 (Holland [14]) If A(O) is 0-2-transitive, then A(O) is a divisible

group.

Proof. Let g E A(O) and n E N. Since gn and g have the same landscape,
</J-l gn</J = 9 for some </J E A(O). Then (</J-l g</J)n = g.

Corollary 4.4 (Holland [14]) Every £-group can be embedded in a divisible £-

group.

Outline of proof. Let G be an £-group. Use the Holland Representation Theorem

to embed G in some A(Ot}. Then embed 0 1 in some 0 for which A(O) is 0-2-
transitive, and do this in a way that embeds A(OI) in A(O).
38 Stephen H. McCleary

5 The transitive structure theory:

"Reduction" to the o-primitive case
Throughout this section (G, fl) will denote a transitive R-permutation group. With-
out transitivity most of the definitions are valid, but most of the results require
some modification (see Section 8).
A convex congruence of ( G, fl) is simply a congruence C (an equivalence relation
on fl which is respected by G) for which each congruence class Ll is convex. We set
C1 S C2 iff Cl refines C2. The key result of this section is that this partial order on
the set of convex congruences is actually a total order.
Always there are the two improper convex congruences (the trivial congruence
£ whose classes are single points, and the congruence whose only class is fl). In
Example 5 there is also a proper convex congruence; its classes are the local copies
of Z . As usual, a block of (G, fl) is a non-empty Ll <:;:; fl such that if Llg n Ll # 0
(g E G) then Llg = Ll. The classes of a convex congruence are convex blocks (0-
blocks). Conversely, because of transitivity the set 3. = {Llg I9 E G} of translates
of anyone o-block Ll forms a convex congruence (we identify congruences with their
corresponding partitions of fl), and each convex congruence is determined by any
one of its classes. Thus for any fixed a E fl, the set of o-blocks Ll containing a
(ordered by inclusion) is in one-to-one order-preserving correspondence with the set
of convex congruences of (G, fl).
Let Ll be an o-block. Then 3. inherits a natural total order from fl. The (not
necessarily faithful) action of G on 3. given by (Llg)h r--; Ll(gh), h E G, preserves
the order on 3., and the resulting group G of order-preserving permutations of 3.
forms a transitive R-permutation group (G, 3.). G is the quotient of the R-group G
by the R-ideal L(3.) = {g E G I~g = ~ for all ~ E 3.}.

Theorem 501 (Holland [15]) Let (G, fl) be a transitive R-permutation group. Let
a E fl. Then
(1) The collection of o-blocks containing a is totally ordered by inclusion.
(2) The collection of convex congruences of (G, fl) is totally ordered by refine-
ment.
(3) An order-isomorphism from the tower of o-blocks containing a onto the tower
of convex congruences is given by Ll r--; 3..

Proof. We prove (1), which will yield (2) and (3). Suppose that .6. and E are
o-blocks containing a which are incomparable under inclusion. By transitivity we
can pick 9 E G such that ag E E \.6.. This makes ~g = E. Then Llg meets .6. (see
Figure 5), making Llg = Ll and contradicting the fact that ag rj. .6..

There is another tower of interest here. For Ll <:;:; fl, let G!'l. = {g E G I Llg = .6.},
the setwise stabilizer of.6.. In the usual one-to-one correspondence .6. r--; G!'l. between
Lattice-ordered permutation groups 39

A
r---------~~------~1 9
r ~
~~--------~v~-------J

Figure 5: Two a-blocks containing a must be comparable under inclusion.

the set of blocks A containing a and the set of subgroups H containing Get (whose
inverse is H I--> aH), the a-blocks correspond precisely to the convex subgroups.
(Let H be a convex subgroup containing Get, and let a :::; {3 :::; Qh for some h E H.
Pick gl, g2 E G+ such that agl = {3 and {3g2 = ah. Then aglg2 = ah, so that
g192h-1 E Get :;;;: H, making 9192 E H. Since e :::; gl :::; glg2, gl E H, so that
(3 E aH.) This gives one version of the next theorem. But clearly these convex
Gt::,.'s are prime t'-subgroups of G. Hence all convex subgroups containing Get are
prime t'-subgroups as well. We record this in

Theorem 502 (Holland [15]) Let (G, n) be a transitive t'-permutation group. Let
a E n. Then
(1) The collection of convex (prime t'-) subgroups G which contain Get is totally
ordered under inclusion.
(2) An order-isomorphism from the tower of o-blocks containing a onto the tower
in (1) is given by 6 I--> G t::..

For any {3 -=1= a E n, there exists in the tower of o-blocks .6. containing a a
covering pair 6 1 C 6.2 (i.e., there is no other A properly between 6. 1 and 6. 2) such
that {3 E 6. 2 but (3 '/: Ai· Namely, 6. 1 is the u.n.ion of the 6.'s omitting (3 and 6. 2 is
the intersection of the 6's containing {3 as well as a. Then C1 = .6. 1 and C2 = .6. 2
form a covering pair of convex congruences, knmvn as the value Val (a, (3). The tower
of covering pairs of convex congruences of (G, n) is denoted by f = r( G, n). The
covering pair of convex congruences of which a I E f consists "ill be denoted by
CI and C/. It is easily checked that any convex congruence which is not CI or CI
for some I E f must be both the u.n.ion of the CI's below it and the intersection of
n
the CI's above it. To avoid confusion between points of and elements of f, the
letter I lLill be reserved for elements of f, i. e., for values, and lLill be boldfaced.
No,,- let I = (CI,CI) E f, and let a E rt Let D.y = aCI/CI (see Figure 6).
Let GI consist of the order-preserving permutations of nl induced by the action
of G o-C- on the chain nl _ (GI , D.y) is a transitive t'-permutation group, ~nd by the
transitiYlty of (G, n), (GI , nl ) is independent of the choice of Q_ GI is a quotient
of an t'-subgroup of G.
Since CI and CI form a covering pair of convex congruences of (G, D), it follows
40 Stephen H. McCleary

aC'"
__-------------J)L~--------------_)
(
a /3
( I) I

easily that the transitive i-permutation group (G.." a.,) has no proper convex con-
gruences at all, i.e., that (G.." a.,) is o-primitive. (G.." a.,) is called an o-primitive
component of (G, D). r( G, D) is the spine of (G, D), the (G.." D.., )'s are the ribs, and
(r, {( G.." a., ) I 'j' E r}) is the skeleton. In Example 5 the spine has two elements.
The top rib is (A(R), R) and the bottom rib is the regular representation (Z, Z) of
the integers.
We recapitulate the preceding discussion, bearing Figure 6 in mind:

Theorem 5.3 Let (G, D) be a tmnsitive l!-permutation group.

(1) Let'j' = (C.."C..,) be a covering pair in the tower of convex congruences of
(G, D). Let a E D. Then the action of Gae.., on the chain D.., = aC'" IC.., of C..,-
classes contained in aC'" induces a group G.., of order-preserving permutations of
D.." and (G.." a.,) is an o-primitive tmnsitive l!-permutation group.
(2) Let /3 # a E D. Then there exists a unique value 'j' = (C.."C..,) = Val(a,/3) E
reG, D) such that aC'" /3 but not aC..,/3.
In many arguments about (G, D), this serves to "reduce" the question to the
o-primitive case. This puts a premium on understanding o-primitive transitive
l!-permutation groups. We address this issue in Section 7.

6 Wreath products
Can eyery conceivable skeleton (r, {( G.." D..,) II' E r}) actually occur? That is,
given a collection {( G.." D..,) I'j' E r} of o-primitive transitive l!-permutation groups
indexed by an arbitrary chain r, must there exist a transitive l!-permutation group
(G, D) whose skeleton it is? An affirmative answer is given by constructing the
wreath product (w, D) = Wr{ (G.." D,) II' E r} [28].
We consider first the wreath produce (G,o' D,o )Wr( G'l' D"l) of just two factors,
with ,0 < ,I, Thinking of r = ho, ,d as running vertically, take D = h,o x D'll
Lattice-ordered permutation groups 41

ordered lexicographically from the top (i.e., by greatest difference). Define an equiv-
alence relation K on 0 by setting

Identify each K-class with 01'0 in the obvious fashion; and of course the chain of
K-classes is a copy of the chain 01'1' Let W consist of those elements w of A(O)
such that
(1) w respects K,
(2) The permutation of the chain of K-classes (identified with 0 1 ) lies in Gl'll
and
(3) For each K-class Ll, the order-isomorphism induced by w from Ll onto Llw
(each identified with 0 0 ) lies in Gl'o'

Then (W,O) is a transitive €-permutation group whose two ribs are (GI'[] , 01'0) and
(Gl'l,0I'J. Example 5 is actually the wreath product (A(Z),Z)Wr(A(R),R); in
that example it happens that W = A(O).
Now we turn to the general case. In the Cartesian product ~ = TII'EI' O-y, choose
any point and label it O. Let 0 consist of the elements cr E ~ having inversely well
ordered support, i.e., such that b E r Icrb) i= Ob)} is inversely well ordered.
Again order 0 lexicographically from the top.
For each, E r define a pair of equivalence relations on the chain 0 by setting
crK-YT ¢} cr(8) = T(8) for all 8>" and
crK-y, ¢} cr(8) = ,(8) for all 8 ~,.
(In the special case r = bo, 'Yd, KI" = Kl'o = K' the equivalence relation defined
above.) Let W' consist of those elements of A(O) which respect all these equivalence
relations. Then W' is a transitive €-permutation group having the K-y's and K-Y's as
convex congruences. For each wE W', and for each a EO and, E r, w induces an
order-isomorphism WO'K~ from aK-Y /K-y onto (aw)K-Y /KI" which via the obvious
identifications with O-y induces in turn an automorphism of O-y. Finally, let W
consist of those elements w E W' for which WO'JC~ E GI' for all a E 0, , E r. The
wreath product Wr{ (G-y, O-y) I, E r} is the transitive €-permutation group (w, 0).
The covering pairs (CI" C-Y) of convex congruences of (w, 0) are precisely the
pairs (K-y, KI') because of the o-primitivity of the given (G-y, O-y)'s (which matters
only for this feature of (w, 0), not for the construction itself). Moreover the 0-
primitive component (W-y, O-y) is the given (GI" O-y) when we take the component
chain O-y to be QC'Y /C-y and make the obvious identification with the given chain
r2-y. Therefore the skeleton of (W, r2) is the given (r, { (GI" O-y) IT E r} ).
Incidentally, this tells us that every chain r occurs as the chain of positive orbits
of GO' for some transitive €-permutation group (G,O). Simply take
(G,O) = Wr{(A(R),R)-y IT E r},
42 Stephen H. McCleary

the wreath product of r copies of (A(R), R).

Theorem 6,1 (Holland and McCleary [20]) Let (G, 0) be a transitive £-permuta-
tion group with skeleton (r, {( G-y , O-y) IT E r}). Then (G, 0) can be embedded in a
natural way in its wreath product Wr{ (G-y, O-y) II E r}.
The embedding of the given chain 0 into the wreath chain is not in general onto
when r is infinite. We return to this matter at the end of the next section.

7 The classification of o-primitive transitive i-permutation

groups
The a-Primitive Classification Theorem says that every a-primitive transitive £-
permutation group looks very much like one of the first three examples of Section 2.
Recall that a transitive £-permutation group is a-primitive if it has no proper convex
congruences; or equivalently, if it has no proper a-blocks; or again equivalently, if the
point stabilizers Go are maximal convex (prime £-) subgroups of G. (See Theorems
5.1 and 5.2.)
To understand a-primitive groups, and even to give the precise statement of
the Classification Theorem, we need to look at a slight enlargement of our chain
O. For an arbitrary 0, we denote by n the completion of by Dedekind cuts of 0
n
(without endpoints). That is, is formed from 0 by filling in each proper Dedekind
cut (each "hole") of 0 with a point. Since 0 is dense in n, each g E A(O) can be
n
uniquely extended to an order-preserving permutation of (which we also denote
by g). Moreover, w(J V g) = max{wf,wg} and dually, even for w E n\o. Thus we
can regard A(O) to be acting as an £-permutation group on n.
Example 6 (A(Q), Q) is 0-2-transitive, but of course (A(Q), Q) = (A(Q), R) zs
not, having Q as one orbit and the chain of irrationals as the other.

For transitive (G, 0), a non-singleton a-block ~ cannot have an endpoint in n

(either in ~ or in O\~). Necessarily its supremum, denoted by sup~, lies in n\n
(unless ~ = 0). Thus every transitive £-permutation group acting on 0 = R is
a-primitive.
It is easy to see that the first three examples in Section 2 are a-primitive, and
that every 0-2-transitive £-permutation group is actually primitive (in the unordered
sense).
We now offer a redescription of Example 3, the periodic example with G = {g E
A(R) I (a + l)g = ag + 1 Va EO}. Let z E A(R) be translation by 1. Then
G = {g E A(R) I zg = gz}, the centralizer ZA(R)Z in A(R) of z. Vice versa, z
generates ZA(R)G since for any f E ZA(R)G and a E R , af is fixed by Go, whence
it follows easily that f is a power of z.
Lattice-ordered permutation groups 43

Now we need to consider a slightly more general kind of period z, lying in

A(O) but not necessarily in A(!1). A period of a transitive i-permutation group
(G,!1) is an element z E A(O)+, one (hence each) of whose orbits {az n In E Z}
is coterminal in 0, and which generates (as a group) ZA(!1)G. An o-primitive
transitive i-permutation group (G,!1) is called periodically o-primitive if it has a
period z (except that (Z, Z) is not counted as periodically o-primitive). The effect of
any g E G on any interval [a, az) determines its effect "one period up" on [az, az 2 ),
and similarly throughout all of!1.
We now prove that every o-primitive transitive i-permutation group is of one of
the three types we have discussed.

TheoreIn 701 (o-Primitive Classification Theorem, McCleary [30]) Every o-primi-

tive transitive i-permutation group is of one of the following three mutually exclusive
types, and all transitive i-permutation groups of these types are o-primitive:
(1) 0-2-transitive.
(2) The regular representation of a subgroup of the additive reals.
(3) Periodically o-primitive.
To prove the Classification Theorem we need a lemma:

LeIIlIlla 702 Let (G,!1) be an o-primitive transitive i-permutation group, and let
w E O. Then the orbit wG is dense in O. Consequently the action of G on wG is
faithful and (G, wG) is o-primitive, making Gw a maximal prime convex i-subgroup
ofG.
Proof Suppose wG were not dense in O. Necessarily w E O\!1. Set aCT if and
only if there is no wg (g E G) between a and T. Then C would be a proper convex
congruence of (G, !1).

Proof of the Classification Theorem. Let (G,!1) be an o-primitive transitive

i-permutation group not satisfying (1) or (2).
Suppose by way of contradiction that for one (hence every) a E !1, GOt. fixes every
(3 < a. Then GOt. <;;: G(3, making GOt. = G(3 by the maximality of GOt., so that all the
stabilizer subgroups Gs coincide, whence it follows that all are {e}. Thus (G,!1) is
regular (i.e., uniquely transitive). Then G is totally ordered since if ag :::; ah then
l5g ~ l5h for all l5, else gh- 1 Ve would move l5 while fixing a, contradicting regularity.
The o-group G has no proper convex subgroups since GOt. = {e} is a maximal convex
subgroup. This makes G Archimedean since if (in additive notation) 0 < a < b and
na < b for all n E N, then Conv{ma ImE Z} would be a proper convex subgroup
of G. By Holder's Theorem (see Example 2 ff.), we have (2).
Therefore we may pick (3 < a which is moved by GOl (see Figure 7). Our
immediate goal is to show that GOl has a first positive orbit ~l = (~dOl' Let
~ = Conv «(3GOl ) be the orbit of Get containing (3. Since ~ is bounded above, we
44 Stephen H. McCleary

may set f = sup 6. E Pick 9 E G such that j3g = ex, so that ex < fg. Let
n.
6. 1 = Conv((fg)G",), the orbit of Go: containing fg (in its Dedekind completion).
To show that 6. 1 is the first positive orbit of Go:, we need to show that for any
>.. such that ex < >.. < fg, there exists! E G such that (fg)! ~ >... By the
Q

lemma we may pick h E G such that ex < fh ~ >... Since exh- 1 < f = sup.6 and
exg- 1 = j3 E 6., we may pick k E Go. such that (exg- 1 )k ~ exh -1. Now exg- 1 kh ~ ex
and (fg)g-lkh = fkh = fh ~ >.. (since f is fixed by k E G Taking f = g-lkhl\e,
Q ).

we have! E Go. and (fg)! ~ >.., making 6. 1 the first positive orbit of Ga. 6. 1 must
be a long orbit, else D would be discrete and by a-primitivity would in fact be a
copy of Z, yielding (2).

9 9

>.. fg
\,...----....v,.--~) h h ~-------------~v __--JJ
Figure 7: Proof that Ga has a first positive orbit 6. 1

Let w'" = sup(6.r)", E n.

(Were (6. 1 )a unbounded above, putting all points
above ex in the same G",-orbit, (G, D) would be a-2-transitive.) Then Wag = wag
for all 9 E G since (6.r) a 9 = (6. r) ",g. Go. ~ Gw~, and by the m:L'Cimality of G n we
have G a = G WQ Then ex < f.L implies wa < wJ.l; for if exg = f.L for some g E G+\Go;,
then wag = Wag = wJ.l and g E G+\Gz;;".
Define z : D - t n
by exz = wa, a one-to-one order-preserving map from D into
n. For any ex E D and 9 E G we have

Wag = wag, i.e., (exg)z = (exz)g.

Hence z maps D onto a dense subset of n
(see the lemma) and thus has a unique
extension to an order-preserving permutation of n which we also denote by z; and
z E ZA(Ii)G.
Conv {exz n In E Z} is an a-block II of (G, D). For if

then for all p E Z we have

Lattice-ordered permutation groups 45

which lies between o:zn+p and o:zn+1+p. By o-primitivity, II = n and the z-orbit
{o:zn In E Z} is coterminal in n.
Because z centralizes G, every o:zn (n E Z) must be fixed. by Ga., and no element
n
of between o:zn and o:zn+1 can be fixed by Ga. (since this holds for n = 0 because
(o:,o:z) = ..6. 1 ). Therefore the n-interval (o:zn,o:zn+1) constitutes a single Ga.-orbit
..6.n+1' Now let W E ZA(fl)G. Since o:w must be fixed by Ga., o:w = o:zn" for some
na.. In fact na. is independent of 0: since for all g E G we have

Hence W is a power of z. This shows that z is a period of (G,n), making (G,n)

periodically o-primitive. This concludes the proof of the Classification Theorem.

Corollary 7.3 In Lemma 7.2, (G,wG) is of the same o-primitivity type and sub-
type (o-2-transitive (pathological/non-pathological), regular, or periodic) as (G, n).

Proof. Now that we have the Classification Theorem, a quick scan of the several
cases makes it clear that the o-primitivity type of (G,wG) cannot differ from that
of (G, n).

Corollary 7.4 (Lloyd [25] and Holland [15]) If (A(n), n) is transitive and 0-
primitive, it must be non-pathologically 0-2-transitive or regular.

Proof. (A(n), n) cannot be periodically o-primitive. For pick 0: E nand e 1= g E

Ga.. Let 9 be the element of A(n) obtained from g by "depressing" g (i.e., making
it agree with e) off the first positive long orbit ~1 of Ga.. Then 9 doesn't commute
with the period z.
Now suppose (A(n), n) is o-2-transitive (and thus 0-3-transitive by Proposition
2.2). Pick 0:1 < 0:2 < 0:3 E n, and pick h E G which moves 0:2 strictly up but fixes
0:1 and 0:3. Now to obtain a non-identity element of bounded support, let k = h V e
and depress k off the supporting interval of k which contains 0:2.

It is easy to use this notion of "depressing" to check that any regular (A(n), n)
is automatically o-primitive, and thus is the regular representation (G, G) of some
subgroup of the additive reals. One obvious example of a G giving rise to a regular
(A(!1),!1) is G = Z. It is far from clear that there are any dense G's which do
so. However 22 1'0 such groups were discovered by T. Ohkuma [38J These groups are
known as Ohkuma groups.

Corollary 7.5 (Holland [16J and McCleary [30]) Let (G, n) be an o-primitive
transitive f-permutation group. Then G is f-simple (i.e., has no proper f-ideals)
unless (G,!1) is 0-2-transitive and has elements of unbounded support.
46 Stephen H. McCleary

Proof. Consider the case in which (G, n) is periodically a-primitive with period
z. Let H i- {e} be an £-ideal of G, and pick e i- h E H. Replacing h by h V h- 1 ,
we may suppose that e < h. For every w E 0, the density of wG in 0 guarantees
that W is moved by some conjugate hw of h. Pick 0: E n. The supports of the hw's
for which 0: S w S o:z provide an open cover of the compact O-interval [o:,o:z], so
there exists a finite collection hWl , ... ,hwn of these hw's having no common fixed
point w in [0:, a:z]. Let k = hWl V ... V h Wn E H+. By periodicity Tk > T for all
TEO. Now let g E G+. Then akrn 2: azg for some m, else T = sup { akrn I mEN}
would be fixed by k. Now for all fJ E [a,az] we have fJk 7n 2: ak 7n 2: azg 2: fJg, and
n
thus fJk 7n 2: fJg for all fJ E by periodicity. By the convexity of H, we have g E H.
Since G is generated by its positive elements, this establishes that H = G.
The 0-2-transitive case (when all elements of G have bounded support) and the
regular case are considerably easier, and we won't bother to treat them.

Later we shall need the following theorem, whose proof we omit:

Theorem 706 (McCleary [33]) Let (G, n) be periodically a-primitive. Then the
(faithful) action (G cn .6. n ) on any long orbit .6.n of GO' is non-pathologically 0-2-
transitive.

The proof of the Classification Theorem showed that the chain of long orbits .6.rn
of G n is order-isomorphic to Z. Between.6. rn and .6. m+1 lies exactly one element
wm of 0, which mayor may not lie in n. It is easy to extend the proof to show
that for any periodically a-primitive (G, n) either
(a) There exists n E N such that wm E n iff n Im (the Config (n) case), or
(b) wm E n only when m = 0 (the Config ( (0) case).
Of course this configuration is independent of the choice of a. The periodic group in
n
Example 3 has Config (1). In fact, all ofthese configurations can occur with = Q.
It is not hard to show that groups of Config (00) are actually primitive (in
the unordered sense), whereas those of finite Config (n) are not primitive since
{az nk IkE Z} forms a block. Moreover, when G = ZA(S1)Z, the full centralizer in
A(n) of the period z, G is actually simple (as a group) when G has Config(oo), and
for finite Config (n) the proper normal subgroups of G are precisely the subgroups
contained in the center (zl1) of G.

We are now in a position to discriminate among the elements of 0, some of which

are more like elements of n than others. Let (G, n) be a transitive £-permutation
group, not necessarily a-primitive. We define a chain no intermediate between n
and n (and depending on G as well as n). There are two cases:
(1) (G,n) may have a smallest non-trivial convex congruence C, i.e., r(G,n)
may have a smallest element fO = (E,C) = (Cl'o,Cl'o). In this case (G,n) is said to
be locally a-primitive and the Cl'o-classes are its primitive segments. We let no be
Lattice-ordered permutation groups 47

the union of the completions LS. of the primitive segments .6..

(2) If (G, n) has no smallest C, let no consist of those wEn such that w E LS.
for o-blocks .6. from arbitrarily small non-trivial convex congruences C, i.e., for
which w is the intersection of (the completions of) a tower of o-blocks coming from
arbitrarily small non-trivial convex congruences.
In both cases, n ~ no ~ n, and in view of Lemma 7.2, no = {w E I wG is n
dense in n}.
We can now expand on the remark after Theorem 6.1 about the embedding of n
in the wreath product chain. The chain embedding fills in certain holes of n with
single points and others with non-singleton segments which turn out to be a-blocks
of the wreath product. When (G, n) is locally a-primitive, no holes are filled in by
single points; and when it is not, the holes filled in by single points are precisely the
elements of nO\n. In both cases, the (unpleasant) holes filled in by non-singleton
segments are those holes w for which w is the intersection of (the completions of) a
tower of a-blocks not coming from arbitrarily small non-trivial convex congruences.
When reG, n) is well ordered, no holes at all are filled in, and the wreath chain is
just n itself.

8 Arbitrary i-permutation groups: The structure theory

Without transitivity, there seems at first to be little hope of developing anything
much like the transitive structure theory. To consider an extreme case, if G =
{e} then every convex subset .6. ~ n is an o-block and every convex equivalence
relation is a convex congruence of (G, n). Nevertheless, it does turn out to be
possible to extend the transitive structure theory to arbitrary i-permutation groups
in a manner which parallels the transitive theory rather closely [34]. Here we first
give a suggestive sketch of how the general theory relates to the transitive theory.
Despite the lack of precision in that overview, we then manage to give a precise
statement (Theorem 8.1) of the information crucial in applications of the theory,
which combines analogues of part (2) of Theorem 5.3 (the reduction to the 0-
primitive case) and of the a-Primitive Classification Theorem.
The way out of the overabundance of a-blocks is to select certain "natural" ones,
and ignore the others. We omit here the precise definition of "natural" , but mention
that in the transitive case (and in the following example) all o-blocks are naturaL

Example 7 Form (G, n) as follows: Begin with (A(Q), Q). Form n by modifying
Q, replacing each element of Q by a copy of R and each element of Q\Q by a
copy of Z. Take G = A(n), which respects the set of local copies of R and the set
of local copies of Z. Then r( G, n) consists of three elements: a greatest '"Y (with
(G y , fLy) = (A( Q), Q», and under it two incomparable '"Y's, one with (G"Y' fLy) =
(A(R),R) and one with (G"Y,n"Y) = (Z,Z).
48 Stephen H. McCleary

As in the example, the collection of natural a-blocks containing any given et E D

again forms a tower. The collection r( G, D) of covering pairs of natural convex
congruences on appropriate subsets of D forms not a tower but a root system,
meaning that the collection of covering pairs above any given one is totally ordered.
Each pair (~=I- (3 E D has a value Val(et,j3) = 'Y E r(G,D), associated with which
is a (not necessarily transitive) a-primitive component (G" D, ). Here a-primitive
means that (G" D,) has no proper natural convex congruences, or equivalently, no
proper natural a-blocks.
In the transitive case, a combination of the Classification Theorem and part (2) of
Theorem 5.3 tells us that for each pair et =I- (3 E D, there exists an a-block Do (namely
etC') containing et and (3, and a convex congruence C of the action (G,6., Do) such
that etC =I- (3C, and such that the transitive €-permutation group (G" D,) arising
from the action of G,6. on Do/C comes from a short list of well understood types.
This turns out to hold even in the general case. The general a-primitivity types are
those listed in Theorem 8.1, and are hardly any different from the transitive types.
The precision of Theorem 8.1 (which makes no reference to "natural" 0- blocks) is
made possible by this listing.
An €-permutation group (G, D) is called transitively deTived if it is obtained from
an a-primitive transitive €-permutation group (G, E), with E dense in itself (i.e.,
with (G, E) not isomorphic to (Z, Z)), by selecting any set of orbits o-G of holes 0-
of E and augmenting E by filling in each hole in each selected orbit with a single
point. (The action of any g EGis unambiguously determined on t.) For example,
(A(Q), Q) is transitively derived, with Q and Q\Q as the orbits.
The transitively derived €-permutation groups themselves split into three types:
0-2-transitively derived (pathological/non-pathological), regularly derived, and pe-
riodically derived. A transitively derived €-permutation group can be derived from
anyone of its orbits, and the type is independent of which orbit of (G, D) we
take to be E (see Lemma 7.2 and Corollary 7.3). Clearly a transitively derived
€-permutation group has no proper a-blocks, natural or otherwise.
(G, D) is called integral if (up to isomorphism) it is obtained from (Z, Z) by
f----
choosing a chain IT and taking D to be IT x Z (i.e., each point in Z is replaced by
a copy of IT), with the action on IT of each integer translation t given simply by
('if, z)t = (rr, zt). For example, if II is a 2-point chain, then (G, D) is isomorphic to
the £I-permutation group consisting of the chain Z with the group of even transla-
tions. Note that the regular transitive €-permutation group (Z, Z) is classified as
integral rather than as regularly derived.
Finally, (G, D) is called static if G = {e}.

Theorem 801 (McCleary [34]) Let (G, D) be an aTbitraTY e-peTmutation group.

Let et =I- (3 E D. Then there exists an a-block Do containing et and (3, and a convex
congTUence C of (G,6., Do) such that etC 1'= (3C, and such that the e-peTmutation group
arising from the action of G,6. on the chain Do/C of C-classes is of one of the following
Lattice-ordered permutation groups 49

jive mutually exclusive types:

(1) Transitively derived:
(a) 0-2- Transitively derived.
(b) Regularly derived.
(c) Periodically derived.
(2) Integral.
(3) Static.
When a and f3 lie in the same G-orbit, .6. and C are uniquely determined.
Now we relate Theorem 8.1 to our discussion of o-primitive components. When a
and f3 lie in the same G-orbit, the uniquely determined .6. and C are both "natural" ,
and the (-permutation group arising from the action of GL). on .6./C is an o-primitive
component (G-y,O-y) of (G,O). Moreover, all non-static a-primitive components
arise in this way.

9 Ordered permutation groups

As promised in the introduction, we pause to indicate the main structure theoretic
results which apply more generally to ordered permutation groups (G, 0), with G
not required to be a sublattice of A(O), or which at least apply to coherent 0-
permutation groups. (Recall that (G,O) is coherent if whenever a < f3 for some
9 E G then ah = f3 for some h E G+.) The orbits of GOt need not be convex.
Accordingly, the focus is instead on their convexifications Conv(wGoJ, which are
known as orbitals. If W2 E Conv(WI G oJ then Conv(w 2 G Ot) = Conv(WI GOt), so the
set of orbitals inherits a total order from O.
Theorem 5.1 holds for ordered permutation groups, and Theorem 5.2 (for convex
subgroups which contain GOt) holds for coherent permutation groups. Theorems
5.3 and 6.1 hold for ordered permutation groups. The o-Primitive Classification
Theorem holds for coherent permutation groups provided that 0-2-transitivity is
relaxed to require not that GOt have only one positive orbit (a requirement equivalent
to o-2-transitivity), but that it have only one positive orbital. Finally, Theorem 8.1
also holds with a similar modification.

10 Application to normal-valued i-groups

In the study of abstract f-groups, the class N of f-groups satisfying
/g :s: g2 P V/,g E G+
plays a very important role. N is a variety (equational class) of (-groups since (*)
can be rewritten as a universally quantified equation "w = e" for some f-group
50 Stephen H. McCleary

word w (constructed from the lattice operations as well as the group operations),
specifically as

An £-group lying in N is called normal-valued. (This term stems from another

characterization of N which makes reference to the notion of a "value" of an abstract
£-group, a different usage of the word "value" from that in £-permutation groups.)
Obviously all abelian £-groups are normal-valued. A(R) is not, as can be seen
by choosing f, 9 E A(R)+ in accord with Figure 8. The same holds for every 0-2-
transitive (C, 0) because of Proposition 2.2. Our goal here is to illustrate how the
structure theory is used by seeing how membership of C in N is reflected in the
structure of (C, 0). We deal first with the transitive case.

f f f

~R
9 9 9

Figure 8: Failure of the identity (*) in A(R)

Theorem 1001 (J. A. Read [39]) Let (C,O) be a transitive £-permutation group.
Then C is normal-valued if and only if all o-primitive components of (C, 0) are
regular.

Proof ( ~). Let (CI" 01') be an o-primitive component of (C, 0), with 01' =
o:CI' / CI' (see Figure 6).
If (CI" 01') is 0-2-transitive, then there exist elements j, 9 E ct
re-creating
Figure 8 (with R replaced by f2-y and II by o:CI'). Moreover, j and 9 must be
induced by some f and 9 in C aC'" and by replacing f by f V e and 9 by g V ewe
may assume that f, g E C!C"f ~ C+. Since

then certainly

Hence G tJ N.
Lattice-ordered permutation groups 51

If (G-y, f2-y) is periodically o-primitive, then Theorem 7.6 guarantees that Figure
8 can be re-created within (G-y, f2-y ), specifically within the action of a stabilizer
subgroup (G-y )""C., on anyone of its long orbits. Then just as in the 0-2-transitive
case, G ¢:.N.
Proof(~). Suppose all o-primitive components of (G,f2) are regular. Suppose
by way of contradiction that for some a E f2 and f, 9 E G+ we have

Let 'Y = Val(afg, a). Let j and 9 be the elements of Gt induced on f2-y = aC-Y jC-y
by f and g. Then

But this is impossible since (G-y, f2-y) is the regular representation of a subgroup of
the additive reals, and j and 9 are non-negative translations of f2-y.

Thanks to Theorem 8.1, the preceding theorem and proof carryover with hardly
any change to intransitive i-permutation groups (since Val(afg, a) cannot be static):

Theorem 1002 Let (G, f2) be an arbitrary i-permutation group. Then G is normal-
valued if and only if all o-primitive components of (G, f2) are regularly derived,
integral, or static.
From the point of view of Theorems 10.1 and 10.2, relatively few i-groups are
normal-valued. Thus it comes as a surprise to discover that

Theorem 1003 (Holland [18]) N is the largest proper variety of i-groups.

Proof. Let w{x) = w(Xl, .. . ,xn ) be an i-group word. Suppose w{h) i=- e for
some substitution in some C-group H. We must show that this happens in every
C-group G which is not normal-valued.
Always w{x) can be written (though far from uniquely) in the standard form

w{x) = VAWij(X) = V
i j i
All Xijk,
j k

a finite sup of finite infs of reduced group words Wij(X), with Xijk = x~l. This is
due to the distributivity of the lattices underlying C-groups and a few other related
C-group properties.
Use the Holland Representation Theorem to view H as an C-permutation group
on some chain~. We are given that there exist CT E ~ and hI, ... ,hn E H such
that
CTw(h) i=- CT, i.e., m~~nCTwij(h) i=- CT.
• J
52 Stephen H. McCleary

Figure 9: crw = max {min {crWll, crW12}, crw21} =I cr

We envision a picture showing how cr is moved successively by the hijle'S in Wij,

with a single picture showing this for all the finitely many Wij'S (see Figure 9).
Now use the Holland Representation Theorem to represent G as an £-permutation
group (G, D). We assume temporarily that (G, D) is transitive. Since G tf. N, The-
orem 10.1 tells us that (G, D) has an o-primitive component (G'Y' ~) which is 0-2-
transitive or periodic. Proposition 2.2 and Theorem 7.6 guarantee that the above
picture can be re-created in (G'Y' D'Y)' (This is because for each hm the picture stip-
ulates the effect of h m at only finitely many points, namely those points that are
tails of hijle-arrOws (when Xijle = xm) or heads of hijle-arrOws (when Xijle = x;;n).
Therefore there exists a E D and gl, ... , gn E G such that for the permutations
91, ... ,9n induced on D'Y = aC'Y jC'Y by gl, ... , gn, we have

That is to say,

Then certainly
aw(g) =I a.
Therefore w(g) =I e, as desired.
The above assumption of transitivity for (G, D) was only for convenience-
because of Theorem 8.1 the proof is essentially the same in the intransitive case.
Alternately, the remarks at the end of Section 3 show that consideration of the
transitive case is enough.

11 £-group laws
In the proof of Theorem 10.3 it was mentioned that the standard form

W(X) = V1\
i j
Wij(X) = VI\ll
i j Ie
Xijle
Lattice-ordered permutation groups 53

is not unique. For example,

(x 1\ e) V (X- l 1\ e) = e

in every i-group (as can be seen by considering i-permutation groups). How can
one determine whether or not this is true of a given w? Equivalently, how can one
determine in the free i-group on {Xl, ... , xn} whether or not the element w(x) is e?
(The usual universal algebraic considerations establish that there is a unique free
i-group of any given rank.)

Theorem 11.1 (Holland and McCleary [21]) Let w be an i-group word. Then
there is an algorithm to decide whether or not 'w = e" is a law of i-groups. That
is, the word problem is solvable in free i-groups.
Proof (semi-formal). First put w in the standard form

w(x) = V/\ Wij(X) = V/\Il Xijk·

i j i j k

Draw a copy of R, and mark the point O. Draw a diagram like Figure 9 (with
o playing the role of (T, and with an arrow for each Xijk, labelling the arrow with

Xm if Xijk = Xm and with x:;;,? if Xijk = x:;;,I. Do this by proceeding in order

through the Xijk'S in anyone Wij, then (restarting at 0) in another Wij, etc. When
a given Xijk-arrOw is drawn, only finitely many points (ends of arrows) have so far
appeared in the diagram. Choose the head of the new Xijk-arrOW to be anyone of
those finitely many points, or choose it to be (any point) in the interval between two
consecutive ones of those points, subject to the following consistency requirement
(where Xijk = x;;I):
For each previously drawn xm-arrow, the head of the new xm-arrow
must bear the same relation (left of, equal to, right of ) the head of the
previous xm-arrow as the tail of the new xm-arrow bears to the tail of
the previous xm-arrow. (Here the reverse of an x~l-arrow is considered
to be an x%l-arrow.)
There are only finitely many possible diagrams, each of which determines a point
7r = maxi minj Owij (see Figure 9). If some 7r i= 0, then "w = e" is not a law of
i-groups-it fails in A(R). Otherwise "w = e" is a law of i-groups because as in
the proof of Theorem 10.3 its failure in an i-permutation group (H, I:) would lead
to a diagram illustrating that failure.

Despite the impression that the preceding proof may have given, it is fairly
practical to use this algorithm by actually drawing the diagrams by hand, at least
when a few refinements are added to cut down on the number of choices to be made.
54 Stephen H. McCleary

12 Big subgroups of doubly transitive A(D)

Our first interpretation of "big" is "maximal". When G = A(D) is 0-2-transitive
(and thus primitive in the unordered sense), the stabilizer subgroups G&. (0: E 0)
are maximal subgroups of G. It seems safe to say that there are no other obvious
maximal subgroups. Any new ones would have to lack fixed points even in 0, and
it can be shown that they would have to contain B(D), the group of elements of
bounded support. Examples have in fact been found by H. D. Macpherson [27] and
by Glass and McCleary [10]. It is not known whether every proper subgroup of an
0-2-transitive A(D) must be contained in a maximal subgroup.
Of course the G&.'s are also maximal prime convex £-subgroups of G. Ball [1]
had earlier produced examples of other maximal prime convex £-subgroups of A(R),
using the Continuum Hypothesis. A curious connection was discovered in [10]:

Theorem 1201 Let A(D) be 0-2-transitive. Then every maximal prime convex £-
subgroup of A(D) is in fact maximal as a subgroup.

Another interpretation of "big" is "small index". For G = A(Q), the stabilizer

subgroups Go: of points 0: E Q have countable index, as do the stabilizer subgroups
of finite sets of points. (Here the pointwise and setwise stabilizer subgroups coin-
cide.)

Theorem 1202 (J. K. Truss [41]) The only proper· subgroups of A(Q) having in-
dex less than 2No are the stabilizer subgroups of finite subsets of Q.

Another nice result about subgroups of A( Q) is

Theorem 1203 (C. Gourion [11]) A(Q) is not the union of any countable tower
of proper subgroups.

13 Closed stabilizer subgroups

Given an £-permutation group (G, D), one would like to be able to characterize the
stabilizer subgroups Go: (0: E D) in £-group language. (In the next section, for
example, we consider the problem of reconstructing the chain D from the £-group
G.) We mentioned in Section 3 that the Ga's are prime convex e-subgroups of G,
and that no more can be said in complete generality. Here we discuss an £-group
property which the stabilizer subgroups satisfy in many important cases, and look
at some other closely related properties. These properties all involve suprema of
arbitrary (infinite) subsets which may happen to exist in G; they are automatically
Lattice-ordered permutation groups 55

valid for finite suprema. All are equivalent to their duals. The presence of these
properties makes an £-permutation group much easier to deal with.
An £-subgroup H of an £-group G is closed in G if whenever g = ViEI hi (g E
G, hi E H), then g E H. We prove below (Theorem 13.2) that the stabilizer
subgroups A(O)et of any (A(O), 0) are closed in A(O).
For any (G, 0) each of the following three conditions implies the next (McCleary
[31]):
(1) Sups in (G,O) are pointwise, i.e., if in G, g = ViE! gi, then for each (3 E 0,
(3g is the sup in 0 of {(3gi liE I}.
(2) (G,O) has closed stabilizer subgroups, i.e., Go: is closed in G for all a E O.
(Of course if Get is closed so is Gf3 for all (3 E aG, so that when (G,O) is transitive
all stabilizer subgroups are closed if anyone of them is.)
(3) The £-group G is completely distributive, i.e.,

1\ Vgij = V 1\ gif(i)
iEI jEJ fEJi iEI

for any collection {gij liE I, j E J} of elements of G for which the indicated
sups and infs exist. (This is a statement about the £-group G, independent of the
particular representation (G,O).)
(1) implies (2) for the same reason that stabilizer subgroups Go: are i-subgroups
of G, and (1) implies (3) since (1) causes the (conditional) complete distributivity
of the chain 0 to carryover to the £-group G. We omit the proof that (2) implies
(3). We mention a few other pleasant facts. Condition (1) for (G, 0) is equivalent
to (1) for (G,O). This is also valid for (2), so that if the stabilizer subgroups of
the points a are closed, so are those of all the cuts &. (1) and (2) are essentially
equivalent (and are equivalent when (G,O) is transitive); and when (G,O) is 0-
primitive (transitive or not), then (1), (2), and (3) are all equivalent. Every £-group
G satisfying (3) has a representation satisfying (1) and (2).
The following theorem is due to Lloyd [26]. Lloyd's proof required a great deal
of ingenuity; the present proof from [34] shows how the structure theory enables us
to obtain the theorem by a routine argument. The lemma helps explain why the
least well understood o-primitive transitive £-permutation groups are those that are
pathologically 0-2-transitive.

LeIllIIla 1301 (McCleary [32]) Let (G, 0) be an o-primitive transitive £-permuta-

tion group. Then (G,O) has closed stabilizer subgroups if and only if (G, 0) is not
pathologically 0-2-transitive.

Proof ( -¢=). Suppose ViEI hi = g (hi EGo" g E G), but that g ¢:. Get, making
a < ag and thus ag- 1 < a. Suppose first that (G,O) is non-pathologically 0-2-
transitive. Pick e < f E G such that supp(J) ~ (ag- 1, a). Then g > f-lg ~ hi for
each i (since when wE supp(J-l) we have wf-lg ~ (ag-1)g = a = ahi ~ whi),
56 Stephen H. McCleary

a contradiction. The periodic case is treated similarly, using Theorem 7.6 and the
periodicity itself. The regular case is trivial since then the stabilizer subgroups are
{e}.
Proof( =}). Omitted.

Because of the close relation of arbitrary o-primitive £-permutation groups to

the transitive ones, Corollary 7.3 makes it clear that Lemma 13.1 (with the excep-
tional case now being "non-pathologically 0-2-transitively derived") doesn't actually
require transitivity.

Theorem 1302 (Lloyd [26]) A(O) has closed stabilizer subgroups.

Proof First we consider the transitive case, phrasing the argument so it will
readily carryover to the intransitive case. Let G = A(O), and let a E O. Suppose
ViEI hi = g (hi EGa, g E G), but that a < ago Let 'i' = Val(a, ag). We may
assume that g and the hi's are the identity on O\<£-Y since modifying them to
make them the identity outside aC-Y leaves them within A(O) and leaves the sup
still valid. Let A be any C.y-class within aC-Y. Then A(A) = {e} (which in the
transitive case forces A to be a singleton). For if not, then by Lemma 7.2 there
exists kEG such that Ak meets the C-y-interval (ag-1C-y,agC-y], and then there
exists e < f E A(Ak), yielding g > r;lg 2 gi, where fI is the extension of f to
o which is the identity off Ak. Thus there is a unique order-isomorphism from A
onto Ag. To obtain a contradiction it would be enough to show that the o-primitive
component (G-y, O-y) has closed stabilizer subgroups. That we already know because
of Corollary 7.4 and Lemma 7.2.
In view of the remark after Lemma 13.1, the argument works even without
transitivity once we observe that (G-y, O-y) is obviously not static and that the
existence of the desired k is clear in the integral case.

Lemma 13.1 helps to shed light on free £-groups. The free £-group Fry of rank 17 >
1 has a representation as a pathologically 0-2-transitive £-permutation group. Such
representations are the main tool for studying free £-groups ([7],[24],[35],[36],[5],[4]).
Indeed, in view of Lemma 13.1 the mere existence of such a representation tells us
that FTf is not completely distributive (as it "shouldn't" be).
Lemma 13.1 can be generalized. (G,O) is said to have the support property if
every non-singleton o-block A supports some e =I g E G.

Theorem 1303 (McCleary [31]) Let (G,O) be a transitive £-permutation group

having the support property. Then (G, 0) has closed stabilizer subgroups if and only
~f (G, 0) is not locally pathologically 0-2-transitive.

Corollary 1304 A wreath product Wr{ (G-y, 0,,) I, E r} of o-primitive transitive

Lattice-ordered permutation groups 57

i-permutation groups has closed stabilizer subgroups unless r has a smallest element
1'0 and (G'Yo'Uyo) is pathologically o-%!-transitive.

14 Reconstruction
To what extent does the i-group A = A(O) determine the chain O? We have seen
that the stabilizer subgroups An (0: EO) are closed prime convex i-subgroups of
A. Conversely, every closed prime convex i-subgroup (# A) of A is a stabilizer
subgroup An (McCleary [29]). This encourages the hope that we might be able to
reconstruct the chain 0, or at least its completion 0, from the collection of closed
prime convex i-subgroups of A(O).
We cannot hope to do this in complete generality since A(O) = {e} for every
well ordered O. Thus we restrict to the transitive case. Even then we cannot
hope to reconstruct 0 itself (as opposed to 0) since clearly A(Q) and A(Q\ Q) are
isomorphic as i-groups via extension to Q = R and restriction to the irrationals
Q\Q. But the reconstruction of 0 works! In fact we can sharpen the conclusion a
bit by cutting 0 down to 0° (defined at the end of Section 7 and not equal to 0
unless (G,O) is o-primitive) because for transitive A(O)

0° = {w E 0 I Aw is a representing subgroup of A}.

Various parts of the next two theorems are due to Lloyd [25], Holland [15], and
McCleary [29].

Theorelll 1401 Let A(Od and A(02) be transitive. Then every i-group isomor-
phism 'lj; : A(Od - t A(02) is induced by some order-isomorphism ¢ : O?
- t og,

meaning that g'lj; = 1>-lg¢ for all g E A(Ol). Hence the pair ('lj;, ¢) gives an i-
permutation group isomorphism .from (A(Od, 0 1 ) onto (A(02),01¢), and Ol¢ is
an orbit of(A(02),Og). In particular, every i-group automorphism of A(OI) isjust
conjugation by some ¢ E A(O?).

Since (A(R), R) has only one orbit and since the two orbits of (A(Q), Q) are
not order-isomorphic, we get the next two corollaries:

Corollary 1402 Every i-group automorphism of A(R) is inner, and similarly for
A(Q).

Corollary 1403 Let (ACO),O) be transitive. If A(O) is isomorphic as an i-group

to A(R), then 0 is order-isomorphic to R; and if A(O) is isomorphic to A(Q), then
o is order-isomorphic to either Q or R\Q.
58 Stephen H. McCleary

For an 0-2-transitive A(O) to have an outer f-group automorphism, it is neces-

sary and sufficient that there be an orbit i= 0 of (A(O), 0) which is order-isomorphic
to O. It is not at all obvious that any such O's exist, but in fact they do-an example
has been constructed by Holland [17].
For the locally o-primitive case, we can relax these hypotheses considerably. The
next theorem says roughly that if an f-group G has two faithful representations as
transitive f-permutation groups, with one of them fairly decent and the other almost
arbitrary, the two representations must be exceedingly much alike.

Theorem 14,4 Let (G 1 , Od be a transitive f-permutation group which is locally

o-primitive but not locally pathologically 0-2-transitive and which has the support
property. Let (G2' O2 ) be any transitive f-permutation group such that G 2 does not
act faithfully on any non-trivial convex congruence of (G2, 02). Then every f-group
isomorphism't/J : G 1 -+ G 2 is induced by some order-isomorphism ¢ : O~ -+ og.
Hence the pair ('lj;, ¢) gives an f-permutation group isomorphism from (G 1 , 0 1 ) onto
(G2,01¢), and 01¢ is an orbit of (G2' og).

We mention that the slightly cleaner statement of the preceding theorem, in

comparison with [8, Theorem 7A], is enabled by the fact that if (G, 0) is a transitive
C-permutation group and wE 0°, then (wG)O, formed from (G,wG), coincides with
the original 0°.
So far we have been dealing with isomorphisms that preserve both the group
structure and the lattice structure. Sometimes preservation of just one of these is
enough!
The history of group isomorphisms involves obtaining conclusions like that of
Theorem 14.1 under ever weaker hypotheses. For example, Corollaries 14.2 and
14.3 hold for group isomorphisms except that in the former the conjugator may
be an order-reversing permutation of R or Q respectively. As it turns out, such
conclusions sometimes hold for subgroups G of A(O) which are not necessarily
sublattices and do not necessarily act transitively on 0:
(G,O) is said to be approximately 0-2-transitive if for all 0::1 < 0::2 and (31 <
(32 E 0 and any open intervals b. 1 containing (31 and b.2 containing (32, there exists
g E G such that O::lg E b. 1 and 0::2g E b. 2 ·

Theorem 14,5 (McCleary and M. Rubin [37]) Let (G 1 , OJ) and (G2, O2) be ap-
proximately 0-2-transitive ordered permutation groups having non-identity elements
of bounded support. Then every group isomorphism 'I/J : G 1 -+ G2 is induced by
some order-isomorphism or anti-order-isomorphism ¢ : 01 -+ rh.

Here is another remarkable theorem involving only the group and not the lattice
(cf. Corollaries 14.2 and 14.3):
Lattice-ordered permutation groups 59

Theorem 1406 (Yu. Gurevich and Holland [12]) There is a sentence in the ele-
mentary language of groups which characterizes A(Q) in the sense that it holds in a
transitive A(O) if and only if 0 is order-isomorphic to Q or to R\Q; and similarly
for A(R).

The other half of the story is due to M. Giraudet [22]:

Theorem 1407 (Giraudet) Let (A(Ol),Ol) and (A(02),02) be 0-2-transitive.

Then every lattice-with-e isomorphism 1jJ : A(Od ---> A(n2) is induced by some
order-isomorphism or anti-order-isomorphism ¢ : fll ---> fl2 (where in the latter
case this is applied to g-l, i.e., g1jJ = ¢-lg-l¢ for all g E G).

Giraudet also established a lattice-with-e analogue of Theorem 14.6.

We have been dealing with the e-group obtained by equipping A(n) with the
pointwise order. How much choice did we really have? That is, how else could
the abstract group A(O) have been lattice-ordered? This leads to a different type
of reconstruction problem, this time of the order (the exact order, not just up to
isomorphism) from the group operation!

Theorem 1408 (Holland [19]) Let A(O) be 0-2-transitive. Then the only lattice-
orderings of the group A(n) yielding an e-group are the pointwise order and its
dual.

And vice versa!

Theorem 1409 (M. Darnel, Giraudet, and McCleary [3]) The only group opera-
tions on the lattice A(R) having the usual identity e are composition and reverse
composition. This holds also for A(Q).

References
[1] R. N. Ball, Full convex e-subgroups of a lattice-ordered group, Ph.D. Thesis,
University of Wisconsin, Madison, Wis., USA, 1974.

[2] R. N. Ball and M. Droste, Normal subgroups of doubly tranBitive automorphism

groups of chains, Trans. Amer. Math. Soc. 290(1985), 647-664.

[3] M. R. Darnel, M. Giraudet, and S. H. McCleary, Uniqueness of the group oper-

ation on the lattice of order-autom071JhisrnB of the real line, Algebra Universalis
33(1995), 419-427.
60 Stephen H. McCleary

[4J M. Droste, Representations of free lattice-ordered groups, Order 10(1993), 375-

381.

[5J M. Droste and S. H. McCleary, The root system of prime subgroups of a free
lattice-ordered group (without GCH), Order 6(1989), 305-309.

[6J M. Droste and S. Shelah, A construction of all normal subgroup lattices of

2-transitive automorphism groups of linearly ordered sets, Israel J. Math.
51(1985), 223-261.

[7J A. M. W. Glass, e-simple lattice-ordered groups, Proc. Edinburgh Math. Soc.

19(1974), 133-138.

[8J -----------, Ordered Permutation Groups, London Math. Soc. Lecture Notes
Series No. 55, Cambridge University Press, Cambridge, England, 1981.

[9J A. M. W. Glass and W. C. Holland (editors), Lattice-ordered groups: Advances

and Techniques, Kluwer Academic Pub., Dordrecht, The Netherlands, 1989.

[10J A. M. W. Glass and S. H. McCleary, Big subgroups of automorphism groups

of doubly homogeneous chains, Ordered Algebraic Structures: The 1991 Con-
rad Conference, ed. J. Martinez and C. Holland, Kluwer Academic Publishers,
Dordrecht, The Netherlands, 1993, 51-71.

[l1J C. Gourion, A propos du groupe des automorphismes de (Q; ::::;), C. R. Acad.

Sci. Paris Ser. I Math. 315(1992), 1329-1331..

[12J Yu. Gurevich and W. C. Holland, Recognizing the real line, Trans. Amer. Math.
Soc. 265(1981), 527-534.

[13J G. Higman, On infinite simple groups, Publ. Math. Debrecen 3(1954), 221-226.

[14J W. C. Holland, The lattice-ordered group of automorphisms of an ordered set,

Michigan Math. J. 1O(1963), 399-408.

[15J -----------, Transitive lattice-ordered permutation groups, Math. Zeit. 87(1965),

420-433.

[16J ___________ , A class of simple lattice-ordered permutation groups, Proc. Amer.

Math. Soc. 16(1965), 326-329.

[17J ___________ , Outer automorphisms of ordered permutation groups, Proc. Edin-

Lattice-ordered permutation groups 61

burgh Math. Soc. 19(1975), 331-344.

[18] __________ , The largest proper variety of lattice-ordered groups, Proc. Amer.
Math. Soc. 57(1976), 25-28.

[19] _______ , Partial orders of the group of automorphisms of the real line, Con-
temporary Math. 131(1992), 197-207.

[20] W. C. Holland and S. H. McCleary, Wreath products of ordered permutation

groups, Pacific J. Math. 31(1969), 703-716.

[21] ___________ , Solvability of the word problem in free lattice-ordered groups, Houston
J. Math. 5(1979), 99-105.

[22] M. (Jambu-)Giraudet, Bi-interpretable groups and lattices, Trans. Amer. Math.

Soc. 278(1983), 253-269.

[23] H. Kneser, Kurvenscharen auf Ringfiiichen, Math. Ann. 91(1924), 135-154.

[24] V. M. Kopytov, Free lattice-ordered groups, Algebra and Logic 18(1979), 259-
270 (English translation).

[25] J. T. Lloyd, Lattice-ordered groups and o-permutation groups, Ph.D. Thesis,

Thlane University, New Orleans, La., USA, 1964.

[26] _________ , Complete distributivity in certain infinite permutation groups, Michi-

gan Math. J. 14(1967), 393-400.

[27] H. D. Macpherson, Large subgroups of infinite symmetric groups, Proc. NATO

AST Conf. on Finite and Infinite Combinatorics, Banff, Alberta, Canada, 1991.

[28] S. H. McCleary, Orbit configurations of ordered permutation groups, Ph.D. The-

sis, University of Wisconsin, Madison, Wis., USA, 1967.

[29] _________ , The closed prime subgroups of certain ordered permutation groups,
Pacific J. Math. 31(1969), 745-753.

[30] ___________ , O-primitive ordered permutation groups, Pacific J. Math. 40(1972),

349-372.

[31] _______ , Closed subgroups of lattice-ordered permutation groups, Trans. Amer.

Math. Soc. 173(1972), 303-314.
62 Stephen H. McCleary

[32] __________ , 0-2-transitive ordered permutation groups, Pacific J. Math.

49(1973), 425-429.

[33] ___________ , O-primitive ordered permutation groups II, Pacific J. Math.

49(1973), 431-443.

[34] ___________ , The structure of intransitive ordered permutation groups, Algebra

Universalis 6(1976), 229-255.

[35] ___________ , Free lattice-ordered groups represented as 0-2-transitive e-

permutation gmups, Trans. Amer. Math. Soc. 290(1985),69-79.

[36] ___________ , An even better representation for free lattice-ordered groups, Trans.
Amer. Math. Soc. 290(1985), 81-100.

[37] S. H. McCleary and M. Rubin, Locally moving groups and the reconstruction
problem for chains and circles, in preparation.

[38] T. Ohkuma, Sur quelques ensembles ordonnes lineairement, Fund. Math.

43(1955), 326-337.

[39] J. A. Read, Wreath products of nonoverlapping lattice-ordered groups, Canad.

Math. Bull. 17(1975), 713-722.

[40] J. Schreier and S. Ulam, Eine Bemerkung iiber d'ie Gruppe der topologischen
Abbildungen der Kreislinie auf sich selbst, Studia Math. 5(1935), 155-159.

[41] J. K. Truss, Infinite permutation groups: subgroups of small index, J. Algebra

120(1989), 495-515.
 On Recovering Structures from Quotients of their
Automorphism Groups

J. K. Truss

Department of Pure Mathematics

University of Leeds
Leeds LS2 9JT
England

1 Introduction
The question as to what information about a structure is provided by knowledge of
its automorphism group has been considered in many cases. In those which I am
principally interested in here, the relevant work was carried out in [10] and [17] (for
symmetric groups), and [7] (for ordered permutation groups). In addition I consider
the rather harder matter of how much we can find out if we are only given suitable
quotients of the automorphism group. In the first of these two cases this is related
to De Bruijn's question on embeddings of quotients of infinite permutation groups.
If JI. is an cardinal (finite or infinite), I write S(JI.) for the full symmetric group
on JI., that is to say, the group of all permutations of JI. (1 - 1 mappings from JI.
onto JI. under function composition). It is well known (see [16] for instance) that,
apart from the alternating group A(JI.) of even permutations, the non-trivial normal
subgroups of S(ji» all have the form S,,(ji» = {g E S(ji» : Isupp gl < I\:} for some
infinite cardinalI\:, where the support of 9 E S(JI.), supp g, is the set of elements
moved by g. Subsets of JI. of cardinality less than I\: are called small.
Two issues which we may address in this connection are the following. In the
first place, can we distinguish the quotient groups S)..(ji»/S,,(JI.) for different values
of 1\:, A, and ji>? This question was posed for instance in [16]. Two obvious senses
of 'distinguish' here are 'up to isomorphism', and 'up to elementary equivalence'.
The sense used in [16] was the former. (In [4] it was initially stated that this
problem was still open, but then it was remarked in proof that the groups can
be distinguished up to isomorphism if we assume the CCH.) As it turns out, a
solution to this problem in a large number of cases had already been included in
[11] Theorem 4.3, which gives a suitable first order bi-interpretability result in the
case where A = ji>+ and c!( 1\:) > 2No, covering both the isomorphism and elementary
equivalence problems. In [3] De Bruijn asked as a related question for which I\: and

63
W. C. Holland (ed.), Ordered Groups and Infinite Permutation Groups, 63-95.
© 1996 Kluwer Academic Publishers.
64 J. K. Truss

fl., 8(/1.)/8" (fl) is embeddabl~ in 8(fl), and results about this are presented in [4].
Th~ second one concerns outer automorphisms. In [J] it is shown that the outer
automorphism group of 8(w)/8w (w) is infinite cyclic, where a typical outer auto-
morphism is induced by the map n f-+ n + 1. The method used there incorporates a
second order int~rpretation of the relevant ring of sets in the quotient group. One of
the by-products of [19] is another proof (in addition to [11]) that this can actually
be done in a first order fashion, and indeed this had already been shown in [11] The-
orem 4.2. A similar result was proved for quoti~nts of order-preserving permutation
groups in [5). It was suggested by M. Droste and M. Rubin that a better way to
view these results was as reconstructions of the tazlof a suitable structure from the
quotient group, which would also imply the outer automorphism result, but which
would cast it ill a more natural s~tting; and that is the viewpoint adopt~d h~re.
Th~ key idea in all the work d~scribed in this paper is to try to int~rpret (in
a precise sense) as much as possible of the underlying structure on which a group
acts in the group itself. To set everything in context I begin by surveying how this
goes for symmetric groups (before passin).'; to quotients), and ordered permutation
groups. For symmetric groups this is essentially classical, at allY rate in the finite
case, and the technique is d~scrib~d in many standard textbooks ([8) for instance).
But I shall emphasize the first order nature of the interpretation, as described for
example in [10]. For ordered permutation groups, an elegant r~sult of Gurevich
and Holland [7) tells us that there is a first order sentence of the language of group
theory which holds in the (full) automorphism group of a doubly homogeneous
chain if and only if that chain is isomorphic to the real numbers R. The idea of the
proof employs an interpretation of R inside its automorphism group, and following
on from this one can immediately read off what the outer automorphisms are of
Aut(R, <), as is done for instance ill [6].
The notat.ion used is fairly standard. l use K, A, and fI. to stand for cardinals
(usually infinite), and IXI for t.he cardinality of the set X. If g and h are elements
of the group G, fl" st.ands for the conjugat.e h- 1 gh, and h is called a conjugaior of g
to gh Permutatiolls (and sometimes other maps) are written on the right of their
arguments.
Overlines such as x are used to stand for 'tuples' (finite sequences (J:l, ... , x,,)
of the appropriate length). If A and B are first order structures, we say that F is a
(first order) interpretation of B in A if [or some n it maps some a-definable (meaning
definable withollt parameters) subset j) of An onto B, the equivalence relation £0'
on lJ given by aKa' ¢:} F(a) = F(a') is a-definable, and is such that for every
m-ary relation symbol Jl of the lan).';uage of B there is a corresponding formula
'PR in t.he language of A with mn free variables such that for all a1, . .. , am E
An, A F= 'PliCa1 . . . ,am) if and only if B F= E.(F(ad,···,F(am )), with similar
conditions for any function or constant symbols of the language of B. The structures
A and Bare bl-interpretable jf there are interpretations FJ of B in A and F2 of
A in B sllch that. F2FJ is O-definable in A and FJF2 is a-definable in B (under
a natural definition of 'colIlposition' of interpretations). \,ye may also consider
Recovering Structures from Quotients of their Automorphism Groups 65

interpretations with parameters, but do not need them here. Moreover these are
not so useful in studying questions about outer automorphisms, since we would then
need to restrict attention to automorphisms fixing the parameters. There is also
a notion of uniform interpretability of one class of structures in another, meaning
that all the interpretations are given by the same formulae. It will be clear that the
interpretations we describe are uniform over large classes of structures, though we
do not labour the point.
I would particularly like to thank Mati Rubin for introducing me to many of the
ideas discussed here, and also Saharon Shelah, for patiently explaining his methods
as described in §4 (which are more fully detailed in [19]). Some ofthe results I discuss
are either consequences of, or generalized by, work of Rubin's [11, 12, 14], and some
of these issues are also discussed in his paper in this volume. I give citations where
appropriate. My main goal however has been to lead up to and elucidate the results
on outer automorphisms, in the natural style mentioned above, and so I have not
given all the possible intermediate results in full generality.

2 Interpretability Results for Symmetric Groups

In this section I review the results from [10] about what can be interpreted in the
symmetric groups S(II:) for cardinals 11:. Essentially one can interpret the structure
(here just a 'set') inside the group. It is an easy consequence that S(II:) has no outer
automorphisms. Or, according to the viewpoint adopted here, we should rather
say that any isomorphism between two symmetric groups is induced by a bijection
between their underlying sets. Since we wish the methods to apply even to the
finite case, we have to stipulate that II: # 6. (It is well known that S(6) is the only
symmetric group admitting an outer automorphism.)
On the whole we represent the underlying structure by means of involutions,
since these are abundant in symmetric groups, and generate the whole group (in a
small number of steps). Although involutions will still be used for the case of the
quotient groups, things do not work so well, and we have to work quite hard by an
alternative approach (due to Shelah [19]) before they can again be invoked.
Consider a symmetric group S(II:) (where II: may be finite). The idea is that we
may represent the members of II: by families of pairs of transpositions. Specifically
a E II: will be represented by the family of all pairs (y, h) of transpositions for
which {a} = supp g n supp h (so that in the definition of 'interpretation' we shall
take n = 2 and D = {(y, h) : y, h transpositions and Isupp Y n supp hi = I}).
The steps required are to show how we can define (i) Y is a transposition, (ii) YI
and Y2 are transpositions for which ISUPPYI nsuPPY21 = 1, (iii) YI,Y2,h l and h2
are transpositions for which Isupp YI n supp Y21 = Isupp hI n sUPP h21 = 1 and
supp YI n supp Y2 = supp hI n sUPP h2.
For (i) we use the formula
transl(x) : x # 1/\ x 2 = 1/\ (Vy)«xx y )2 = 1 V (XX Y )3 = 1).
66 J. K. Truss

Lemma 2.1 If K 2: ;) /\ K ¥ 1,6 then for any 9 E S( K), S( K) F= iT'ansl (g) if and
only if 9 is a transposition.

Proof. If 9 is a transposition (ap), then so is gh, ('115) say. If {a, j3} =

{'I, 15} then
ggh = =
1. If {a, j3} n b, 15} 0 then ggh has order 2. Otherwise I{ a, j3} n b, 15}1 = 1.
Suppose that a = 15. Then ggh = (aP'Y) , which has order 3.
Conversely suppose that S(K) F= transl(g). Then 9 has order 2, so is a product
of one or more 2-cycles. Suppose for a contradiction that there are at least two,
(ap) and ('115). Since K ¥ 1, there is some (rf. {a,j3,'Y,15}. Letting h = (a P '115 f)
we find that ggh maps a to 'I to f, so it must have order 3, and hence 9 moves
f. Suppose 9 interchanges f and C. Since K ¥ 6, there is some other element 7]
say. This time letting h = (a (3,15 ( CT7) we find that ggh maps a to 'I to f to 7],
contrary to ggh of order 2 or 3.

Note that trans1 is true of(01)(23) in S(4) even though it is not a transposition.
In this case we can use an ad hoc method to distinguish the transpositions:
trans(:r) : transl(x) /\ (:3y)(x Y ¥ x /\ (xx Y )3 = 1).
This is still clearly true of all transpositions (for K 2: 3), hut is not true for the
products of two 2-cycles in S( 1). The group S(6) however remains as an exception,
and this is because there is an automorphism in this case which interchanges (01)
and (01)(23)(45). It follows that there can be no way of defining the class of trans-
positions here, for any definition would have to be preserved by this automorphism.
To characterize when 18upp(g) n supp(h)1 = 1 we use the formula
overlap(x, y) : lrans(x) /\ tnlns{y) /\ x ¥ y /\ (xy)3 = 1.

Lemma 2.2 If K 2: ;{ and K ¥ 6 then for any y, hE S{K),

S(K) F= overlap(y, h)
'If and only if

g, h are transpositions for which 18upp(g) n 8upp(h) = l. 1

Proof. Let 9 = (Ct j3) and h = ('I b). If {a, j3} and b, 15} intersect in a single
point, a = 15 say, then .lilt = (n p)b a) = (a ;3 I), which has order 3. If they are
disjoint, yh has order 2.
Since a pair of overlapping transpositions is meant to 'code' the point of inter-
section, we also need to express when two such intersections are equal. To ease this
we use as an auxiliary formula
tl'iple(a:, y, z) : ovel'lap(x, y) /\ oveT'iap(x, z) /\ oveT'iap(y, z) /\ x Y ¥ z.
In terms of this we can introduce the formula
equal( Xl, Yl, x2, Y2) : o1!f:rlap( Xl, yd/\ overlap( X2, Y2) /\ [( Xl = X2)V
(Xl = Y2) V tl'iplr:(:l:J, ,'1:2, Y2)]/\ [(Yl = X2) V (Yl = Y2) V il'iplc(Yl' X2, Y2)].
Recovering Structures from Quotients of their Automorphism Groups 67

LeIIlIIla 2.3 If /'i, 2: 3 and /'i, "# 6 then for any g, h, k E S(/'i,),

S(",) 1= triple(g, h, k)
if and only if there are distinct a, {3, 1, 6 E K such that 9 = (a (3), h = (a 1), and
k = (a 6).
Proof. Suppose that S(",) 1= trip/e(g, h, k). By Lemma 2.2 we may write 9 =
(a (3), h =(a 1). If k = ({31) then gh =
k, contrary to hypothesis. So supp 9 n
supp k =supp h n supp k =
{a} as required.

if and only if gi, hi are all transpositions, and there is a E ", such that supp gl n
supp hI =supp g2 n supp h2 {a}. =
Proof If S(K) 1= equal(gl,h l ,g2,h2) we let gl (a{3),hl = =
(al),g2 = (6e),
and h2 = (6(), and suppose for a contradiction that a "# 6. If triple(gl' g2, h2) then
6 E {a,{3} and so 6 = (3. Then trip/e(h l ,g2,h 2) must fail (as otherwise 6 = 1) and
so hI = g2 or h 2 . But each of these is impossible since (3 f/. {a,I}. It follows in
the same way that -,triple(h l , g2, h2)' Since gl f- hI and g2 "# h2 we have either
gl = g2 /\ hI = h2 or gl = h2 A hI = g2, each of which implies a = 6 again.
Conversely suppose that gl = (a{3), hI = (al), g2 = (a6), h2 = (ae) where
(3 f- 1 and 6 "# t. Then overlap(gl, hI) and overlap(g2, h2)' The cases {3 = 6, e and
{3 f/. {6, c} correspond to gl = g2, h 2, and trip/e(gl, g2, h 2) respectively, and similarly
for the final clause.

Corollary 2.5 If K 2: 3 and K "# 6, then (/'i" =) is interpretable in S(K).

Proof. We let n = 2 and D = {(g, h) : S(K) 1= overlap(g, h)}. Then the map
F is given by F((g, h» =a provided {a} =
supp 9 n supp h. The only relation on
the structure is equality, and the formula equal introduced above serves as in the
definition of 'interpretation' to express when two pairs encode the same point.

TheoreIIl 2.6 Let X and Y be sets of cardinality "# 6, and let () be an isomorphism
SeX) ~ S(Y). Then there is a bijection f : X ~ Y such that for every 9 E
SeX), g8 = f-lg/.

Proof. If X or Y is finite then so is the other, and as IS(X)I = IS(Y)I, IXI =

IY I· This means that we can check the cases IX I = 1, 2 directly. Now assuming
IXI, IYI 2:3 we may use the interpretation given above. Specifically we let af = b
if there are distinct aI, a2 E X - {a} and distinct bl , b2 E Y - {b} such that
(a ai)8 = (b b;) for i =1,2. This is well-defined since if a~, b~ are other choices,
68 J. K. Truss

then by Lemma 2.4, SeX) F equal«a, ad, (a, a2), (a, aD, (a, a~», and as () is an
isomorphism, S(Y) F equal«b, bt), (b, b2 ), (a, aDo, (a, a~)8). By Lemma 2.4 again,
(a aDo = (b bi). Similarly one checks that f is 1 - 1 and onto.
For the last part we start by noting that for distinct a, bE X, (a b)O = (af b!).
For by Lemma 2.1 and the remark after it the class of transpositions is definable,
so that (a b)O = (e d) for some e,d E Y. It suffices to show that af E {e,d}.
Choose bl EX - {a,b}. By Lemma 2.2, SeX) F overlap«ab),(ab l )) and so
S(Y) F overlap«a b)O, (a b')O). Hence supp«a b)O) n supp«a b')8) has just one
member, and by definition of f, this element, which lies in {e, d}, is equal to af.
Therefore if 9 E SeX) and a f. b, (agf bg!) = (ag bg)O = (g-l(a b)g)O =
(g-1)6(a b)Ol = (l)-l(af bf)(l) = (afl bfl). Letting a,b,e be any three dis-
tinct members of X, agf is represented in S(Y) by the pair «agf bg!), (agf eg!))
and so also by «afl bfgS), (afl efl)), and it follows that agf = afl. As a
was arbitrary, gf = fgo so that l =f-lgf as required.

3 Automorphism Groups of Chains

In this section I consider automorphism groups of chains, that is to say, linearly

ordered sets. If (0, <) is a chain, we write A(O) for the group of all order-
automorphisms of 0, as in [6], and we write S1 for the order-completion of O. Then
A(O) may be regarded as a subgroup of A(S1) in the natural way. This is useful for
various purposes; for instance it enables us to refer to arbitrary 'convex' subsets of
o as 'intervals', even though their endpoints may not lie in 0 itself; but we do have
to keep track of where a member of A(O) is being viewed as acting.
It is usually easiest to restrict attention to doubly homogeneous chains, which by
definition are those such that for any Xl < X2 and Yl < Y2 there is 9 E A(O) such
that Xlg = Yl and x2g= Y2. Much of the reconstruction work can be carried out
more generally, but the details are considerably more complicated, and from now on
I shall therefore assume 0 is doubly homogeneous. Provided this holds, there is an
elegant characterization due to Holland of when two elements of A(O) are conjugate
(see [6]), in terms of the 'orbitals' of these elements. If 9 E A(O) and x E 0 let us say
that the orbital of 9 containing x is the subset {y : (3m, n E Z)( xgm :'S Y :'S xgn)}.
Thus an orbital is a convex subset 'spanned' by an orbit. For any orbital X of g, 9
is the identity on X (in which case IXI = 1), or it is strictly increasing, x < xg, for
all x EX, or is strictly decreasing, xg < X for all X EX. We say that the parity of
the orbital is 0, + 1, -1 in these three cases respectively, and the orbitals of parity
± 1 are called non-trivial. It is clear that the union of all non-trivial orbitals of 9 is
equal to its support. The family of all orbitals of 9 is itself linearly ordered in the
obvious way, and the orbital pattern of 9 is the ordered set of orbitals together with
the function assigning to each orbital its parity. Holland's criterion then says that
(for doubly homogeneous 0) two elements of A(O) are conjugate if and only if they
have isomorphic orbital patterns.
 Recovering Structures from Quotients of their Automorphism Groups 69

We let the cofinality (coinitiality) of 0 be the least cardinality of a cofinal (re-

spectively coinitial) subset, and its coterminality is the greater of the two.
It is rather easy to see that in all cases A(O) has three normal subgroups, usually
written L(O), R(O), and B(O), and defined by

L(O) = {g E A(O) : supp 9 is bounded above},

R(O) = {g E A(O) : supp 9 is bounded below},
and

B(O) = L(O) n R(O) = {g E A(O) : supp 9 is bounded above and below}.

In the most familiar examples, (0 = Q or R) these are the only proper non-trivial
normal subgroups, but for general 0 (for instance the 'long line') there will be many
others. See [2] for instance.
For certain chains, very strong interpretability results are known. Specifically,
in [7] it is shown that there is are sentences r.p and tP such that if 0 is a doubly
homogeneous chain, then A(O) F r.p if and only if 0 ~ Rand A(O) F tP if and only
if 0 ~ Q or fr (the irrationals). The proof given in [7] goes via the elegant theory
of lattice-ordered groups. For my purposes it is easier to stick to the pure language
of group theory, and indicate how we can interpret many properties inside A(O),
sufficiently well to be able to recover Rand Q or fr.
The idea of the method presented here is to try to represent 0 inside A(O) by
means of carefully chosen elements. The first attempt is to use elements which have
just one non-trivial orbital, which seem the easiest to handle. These are referred to
in [6] as bumps, in view of the shape of their graphs (which deviate from the identity
function in just one interval). In [5] and [20] we used bumps whose support takes
the form (a, 00) for some a E n. Indeed in [5] it was essential to do so, since in the
quotient group A(O)/ L(n) all bumps with bounded support become trivial. The
formula which enables us to capture elements of this type (at any rate for suitable
0), and which was introduced in [20] is

conjl(x) : x -I 1/\ (Vy)(3z)(xx Y =X Z ).

This asserts that x is not 1, and the product of x with any conjugate is again
conjugate to x. More generally one can consider for n :::: 1 a formula conjn(x) which
may be needed in order to make finer distinctions, and which asserts that x is not
1 and the products of x with a conjugate of x fall into exactly n conjugacy classes.
Study of the formula conj1 works well in the case where 0 has countable cofinality,
and it was used in [20] to give an easy way of distinguishing the elementary theories
of A(Q) and A(R). In this case it is rather easy to check that 9 E A(O) satisfies
conj1 if and only ifit has exactly one non-trivial orbital, which is unbounded (above,
70 J. K. Truss

below, or both). Again in [5] the same formula was used to investigate properties
of the corresponding quotient groups, but almost entirely for r2 having countable
coterminality. Since in this section I do not yet want to restrict attention to the case
of countable coterminality, I focus rather on bumps with bounded support. Even
these are really too restrictive, since any non-trivial orbital clearly has coterminality
No, and so any particular bump will only suffice to 'encode' points a of IT which
happen to be limits of countable sequences (either increasing or decreasing). For
the interpretation we instead use (equivalence classes of) elements, not necessarily
bumps, which can 'encode' arbitrary cuts in r2.
I now introduce one by one various formulae of the language of group theory
which will express certain properties of elements of A(r2) (always under the assump-
tion of double homogeneity). The method is a modification of the one described
in [6]. The following is one of the key points (in partially recovering the lattice
ordering of A(r2) inside the pure group).
comp(x) : (3y)(3z)(y -=J. 1/\ z -=J. 1 /\ (Vt)(yzx t = zx t V)~.

Lem.ma 3.1 For any p E A(r2), A(r2) F r:nmp(p) if and only if either p has no
orbitals of parity +1, or it has no orbitals of parity -1.

Proof. Here camp stands for 'comparability' of p with the identity 1, for in the
natural lattice ordering on A(r2), an element is comparable with 1 if and only if it
satisfies the given condition (it is ::; 1 if it has no orbitals of parity + 1 and it is 2: 1
if it has none of parity -1).
Suppose first that p has no orbitals of parity -1 (so it is 'non-decreasing'). We
choose bumps f and 9 with disjoint supports (a, b) and (c, d) (where as remarked
above, the endpoints lie in IT), and such that b < c. This is possible by double
homogeneity. Then (by Holland's criterion, for instance), any conjugate ph of p is
also non-decreasing, and so (r, d)ph is disjoint from (0, b). But sUpp(gph) = (c, d)ph
and it follows that J and gPh corrunute. Similarly if p has no orbitals of parity +1
(this time taking d < a).
Conversely, suppose that p has orbitals of parity +1 and -1, and let f, 9 -=J. 1.
\Ve have to show that 9 has a conjugate gh, where h is conjugate to p, which
does not commute with f. Let (a, b) be non-trivial orbital of f, and (c, d) a non-
trivial orbital of g. By replacing J and/or 9 by their inverses if necessary we may
suppose that these orbitals have parity +1. Pick if possible x E (a, b) and y E (r, d)
such that yg2 < x:. Let q E A(r2) lIlap y to x, yg to xf, and yg2 to x f3. Since
p has an orbital of parity + 1, q may be taken to be a conjllgate ph of p. Then
x:jgP
• h
= ;I:fl]- 1 gl} = xj'3 and a:g Ph f = xq- 1 gq f = J: f~~,so JgP h T-t. gF. . h J . Tf no
choice of J:, y is possible, then Ii ::; c, and a similar argument using an orbital of p
of parity -I is applied.

aporot(x, y) : (3z)(comp(z) /\ z -=J. I 1\ (Vt)(J;yZI = yzt xo».

Recovering Structures from Quotients of their Automorphism Groups 71

LenlIna 3.2 For any f, g E A(O), A(O) F apart(f, g) if and only if either supp(f) <
supp(g) or supp(f) > supp(g) (where supp(f) < supp(g) means x E supp(f) /\ Y E
supp(g) ~ x < y).

Proof. If supp(f) < supp(g) take any 'positive' h, (meaning that it is not equal
to 1, and it has no orbitals of parity -1). Then any conjugate hk of h is also
positive, and so supp(f) < supp(ghk). Thus f and ghk have disjoint supports so
commute. Similarly if supp(f) > supp(g) (but now using negative h).
Conversely suppose neither supp(f) < supp(g) nor supp(f) > supp(g). Take
any h i= 1 and suppose it has an orbital of parity +1. Since supp(f) < supp(g) is
false there are x E supp(f) and y E supp(g) with x 2:: y. As in the previous proof
we assume for ease that x < xf and y < yg. Then there is a conjugate hk of h
k
taking yg-2 to x, yg-I to xf, and y to xf3, and f does not commute with gh since
xfgh k = xf3 but xghkf =
xf2. So A(O) F ..,apart(f,g).

bump(x) : x i= 1/\ ('t/y)('t/z)(apart(y, z) /\ x = yz -+ (y = 1 V z = 1».

Lemma 3.3 For any f E A(O), A(O) F bump(f) if and only if f is a bump.

Proof. If g and h are elements of A(O) with disjoint supports then the family of
non-trivial orbitals of gh is clearly equal to the unions of the corresponding families
for g and h separately. Hence if f is a bump in A(O) it cannot be written as gh
where g, h are non-trivial elements with disjoint supports, and so A(O) F bump(f).
Conversely, if f i= 1 and f is not a bump, it has at least 2 non-trivial orbitals, and
so there is x E IT fixed by f such that each of ( -00, x) n supp(f) and (x, (0) n supp(f)
is non-empty. We let yg = yf if y < x and yh = yf if y > x, and g and h fix all
other points. Then g and h witness ..,bump(f).

bbump(x) : bump(x) /\ (3y)(comp(y) /\ apart(x, x Y ) /\ apart(x, x y - I )).

Lemma 3.4 For any f E A(O), A(O) F bbump(f) if and only if f is a bump with
bounded support.

Proof. If f is a bump with supp(f) ~ (a, b), let al < bi < a < b < a2 < b2 , and
let p be a positive element taking al to a to a2 and bi to b to b2 . This p is a witness
for y, and so A(O) F bbump(f).
Conversely, if bbump(f) and p is a witness for y, we either have supp(fp-I)
< supp(f) < supp(fP) (if p is positive), or supp(fp-I) > supp(f) > supp(fP) (if p
is negative).

orbital(x, y) : bump(x) /\ (3yd(3Y2)(apart(x, yd /\ apart(x, Y2) /\ Y = XYIY2).

Lemma 3.5 For any g, h E A(O), A(O) F orbital(g, h) if and only if g is the
restriction of h to one of its non-trivial orbitals.
72 J. K. Truss

Proof. Here by saying that 9 is a restriction of h to X we mean that for every

x E X, xg = xh, and for all x rt. X, xg= x. If 9 is a restriction of h to an orbit.al
(a, b) of h we let. hI, h2 be the restrictions of h to (-=, a) and (b, =) respectively.
Then hl,h2 are witnesses for VI and V2.
Conversely, if orbital(g, h) is t.rne, and h1' h2 are witnesses for VI and V2, then
Sllpp(g) must be disjoint from supp(hl) U SUpp(h2)' and 9 must be equal to the
restriction of h to its support. Since 9 is a bump, it is the restriction of h to one of
its orbitals.

\Ve introduce some further formulae, alld state their properties without proof in
a lemma.

disj(x, V) : (Vz)(Vt)(orbital(z, x) /\ ol'bital(v, t) ---> apal't(z, i)).

dcnse(x) : (VV)(disj(x, V) --+ V = 1).
half(x) : (::ly)(y I- 1/\ Clpart(x, y) /\ dense(xy)).
end(x, y) : (3z)(z I- 1/\ apart(x, z) /\ aparl(y, z) /\ dense(xz) /\ dense(vz)).

Lemma 3.6 (I) FaT' any g, hE A(n), A(n) F disj(g, h) if and only if 9 and h have
dii'iJoint supports.
(ii) For any 9 E A(n), A(n) F dense(g) if and only If supp(g) is dense in n.
(iii) For any 9 E A(n), A(n) F half(g) if and only if for some a E IT, supp(g)
is a dense subset of(a,IXJ) or (-=,a).
(iv) For any g, h E A(n), A(O) F end(g, h) if and only if for some a E D,
supp(g) and supp(h) are both dense subsets of (a, =) or arf: both dense subsets of
(-=,0)

Lemma 3.7 For any a E D there are g, h E A(n) such that supp(g) is a dense
subset of (a, =) and supp(h) is a dense subset of (-=, a). Moreover g, h may be
chosen positive (or negative).

Pmof For any x E (b, c) ~ (a, 'XI) with b < c there is a bump with support
a subset of (b, c) and containing x, as follows by double homogeneity. By Zorn's
Lemma we choose a maximal pairwise disjoint family of open intervals which are
supports of bumps of this form, and let 9 be chosen having the members of this
family as non-trivial orbitals. Similarly for h.

These results have served to indicate how we may attempt to interpret n, or

more accurately, D, inside A(n). Since A(O) has an outer automorphism, at any
rate in the cases where n is isomorphic to its reversed chain n* (such as Q and
R), we cannot hope for this as it stands. This is one reason why in [6] and [7] the
lattice-ordered group language is used. However we may find an interpretation with
one parameter (that is by expanding the language to 'tell left from right').

Theorem 3.8 FaT any doubly homogeneous chain n, D is interpretable zn (A(n), g)

where g is a suitable member of A(n).
Recovering Structures from Quotients of their Automorphism Groups 73

Proof. Choose a E 0 and let 9 be a member of A(O) whose support is dense

in (a, 00). In the definition of 'interpretation' we take n = 1 and D to be the set
of elements of A(O) whose support is dense in (b, 00) for some b E O. Then D
is definable (without parameters) in (A(O), g) since h E D if and only if A(O) F
halJ(h) 1\ 3k 13k 2 (subset(gk 1 , h) 1\ subset(hk2, g)) where subset(x, y) is a formula
expressing 'supp(x) ~ supp(y)' (for instance (\fz)(disj(y, z) --t disj(x, z))). If we
let F(h) = b provided supp(h) is dense in (b, 00), the fact that F is surjective follows
from Lemma 3.7, and the definability of 'equality' and < uses the formulae end and
subset.

One would really like to interpret 0 itself in (A(O), g), but this is not possible
unless 0 is order-complete (when it is trivial), since then A(O) = A(O - 0) (under
the natural identification). This is the reason for the difference in the results of [7]
for Rand Q. In the latter case we can only 'recognize' Q up to complementation
in R. We remark that the results described here for doubly homogeneous chains
have been extensively generalized in (at least) two directions by Rubin, in [13]
for topological spaces (where the results are phrased in terms of 'locally movable
groups'; see Item 11 in the list of faithful classes; also see his paper [15] in this
volume), and for 'trees' in [14].
Here I just state the consequences of the development so far for outer automor-
phisms.

Theorem 3.9 Suppose that (0 1 , <) and (0 2 , <) are doubly homogeneous chains
and that e is an isomorphism from A(Od to A(02). Then there is a bijection J
from 0 1 to O2 , which is either order-preserving or order-reversing, and such that
Jor every 9 E A(Od,l = f-1 g J. Moreover J takes 0 1 to an orbit oj A(02) in O 2 •

Proof. We let af = b for a E 0 1 , bE O2 , if there are 9 E A(Od, hE A(02) such

that supp(g) is dense in (a, 00) and h is dense in (b, 00) or (-00, b), and h = l . The
fact that this definition is independent of the choices of 9 and h follows from the
previous lemmas. Moreover it is easy to check that if supp(h) ~ (b,oo) «-oo,b»
for some b and choices of 9 and h, then this holds in all cases, and then J is order-
preserving ( order-reversing, respectively), and in either instance J is 1 - l.
Finally we have to show that for each g E A(Od, l = f-1 g f. For ease we
suppose that J is order-preserving. Pick any a E 0 1 and pick gl with support
dense in (a, 00). Then by definition of f, gf has support dense in (af, 00), and as
supp(gf) = (SUPP(gl»g, which is dense in (ag, 00), similarly supp( (gf)6) is dense in
(agJ, 00). But (gf)6 = (gOCg"), and as supp(gO is dense in (af,oo),supp«gnc g"))
is dense in (afl, 00). We deduce that agf = afl. As a was arbitrary, gf = f g6
and so l = J-1 g f.
For the last part, let a E 0 1 and 9 E A(Ol)' Then (ag)f = (af)g! = (af)l,
so as g6 E A(02), aJ and (ag)f are in the same A(02)-orbit of O2 • Since the
74 J. K. Truss

argument can be reversed it follows that the image of Dl under f is equal to an

orbit of IT2 under the action of A(D2)'

Corollary 3.10 (i) If D is doubly homogeneous, oT'der-complete, and n =: n* (the

T'eversed chain) then the outeT'-automoT'phism gmup of A(n) has order 2.
(il) If D IS oT'der-complete and D ~ n* then A(D) has no outeT' automoT'phisms.

This gives as an example of a doubly homogeneous chain wit.h no outer auto-

morphisms t.he 'right long line' L + (see [5)). To obtain corresponding results for
D which are not complete, it appears that we would need to strengthen the inter-
pretability result. In cases where we can show that n itself is interpretable in A(D),
the analogue of this corollary will apply. For D = Q this is given in [6].

4 Interpretability in SA (/1) / Sf> (/1)

In this section I show how it is possible to interpret IIlany 'set-theoretical' properties
inside 5 A(J.l)/ 5" (J.l), by representing subsets of 11 via support.s of suitably chosen
elements (always up to less than '/'C mistakes'), and consequently to interpret the
ring of suhsets of J.l of cardinality less than ,\ modulo those of cardinality less than
/'C. This work was originally carried out in a rather more general setting in [11]. The
method presented here is due to Shelah, where the key idea is to use sequences whose
entries are transitive representations of a specific finite non-commutative group to
represent the subsets, which enables us to capture disjointness of their supports via
a coIlllllutativity condition. To make things work more smoothly, it helps if the
group used is simple, as this ensures that its action will be either faithful or trivial,
and the number of cases to be considered is reduced. Any non-abelian finite simple
group would probably do, but we fix on the alternating group A(5) for definiteness.
(In fact at one point in the argument we use the fact that the outer automorphism
group of A(5) is known (it is just 5(5)). We introduce the necessary formulae one
by one, and outline why they represent what is required. Fuller details are given in
[19].
If ?J is a tuple of elements of a group, we let (g) be the subgroup generated by
the set of its entries, and supp(?J) be the union of their supports. If 7 and?J are
tuples of the same length n then 7·?J is the pointwise product (hYl,"" /"Yn)'
Let G be a fixed finite group, and let G = (aI, a2,.·., an) be an enumeration of G
(with al = the identity, for ease). Then we let diag(G, x) be the conjunction over
all i, j, k between 1 and 11 for which aiaj = ak of the formulae XiXj = Xk. This is
intended to say that (Xl, J:2, ... , xn) is a 'copy' of G (in the specified enumeration),
but actually just says that it is a homomorphic image.

Lemma 4.1 Snppose 1 E S\(;l) IS such that 5 A (p)/S,,(J.l) 1=

Recovering Structures from Quotients of their Automorphism Groups 75

diag(G, S",(J.l).J). Then there is a small union X oJ (I)-orbits such that iJa E J.l-X,
and aiaj = ak, then aJ;Ji =
ah·

This is plausible, but not quite obvious in the case where K; = w, since it might
appear that we have to close up the set of exceptional points under the action of
(f). The proof is given in [19].

LeIllIlla 4.2 For any1, S>.(J.l)/S",(J.l) F= diag(A(5), S",(J.l).f) iJ and only iJ there is
a small union X oj orbits oj (f) on J.l such that Jor every orbit Y oj (f) on J.l- X,
the action oj (f) on Y is isomorphic to some transitive action oj A(5) (so that
IYI = 1,5,6,10,12,15,20,30, or 60, and then we say that 1 acts as A(5) with this
degree).

Proof. This is immediate from Lemma 4.1 on remarking that for orbits Youtside
J.l - X, the action of (f) on Y precisely corresponds to some transitive action of
A(5). The fact that the possible values of IYI are as stated follows from the fact that
any transitive action of A(5) is isomorphic to its action on a coset space [A(5) : H]
for some subgroup H of A(5), and the possible orders of subgroups of A(5) are
1,2,3,4,5,6,10,12, and 60.
With this lemma in mind we may define for any 1 E S>.(J.l) of length 60 the
cardinals vm(1) for m E {I, 5, 6,10,12,15,20,30, 50} by vmCl) = the number of
1
(f)-orbits on J.l on which acts as A(5) with degree m. The significant values (that
is, those which are preserved on passing to the coset S>.(J.l).f) are those vmCl) which
are 2: K, and these provide a 'profile' of S,,(J.l).f characterizing it up to conjugacy.
Ideally we would like to restrict attention to those 1 for which vmCl) < K; for
m ~ {I, 5}. Thus the action of (f) on all but a small set of its orbits would be
either trivial or 'natural' (that is, the intended faithful action of degree 5). And
if we are viewing expressing disjointness as the main goal at this stage then the
natural way to express S,,(J.l).f, S,,(J.l).g disjoint would be to say that every entry
of S",(J.l).f commutes with every entry of S",(J.l).g.
Unfortunately neither of these ideas works so straightforwardly, and we have to
go by a longer route. In the first place the formula diag(A(5), x) does not rule out
the other representations. And there are several ways in which S",(J.l).f, S,,(J.l).g
can commute entrywise without IsuppCl) n supp(g) I < K holding. To analyse the
possibilities for this let us focus attention on the commuting actions of 1 and g on
subsets X of K such that (f, g) acts transitively on X and (f), (g) each act on all
their orbits in X as A(5).
The case we really wish to capture is where either] or g acts trivially on X.
Other possibilities however are these: it may be possible to write X = {Xij : i <
m,j < n} so that the orbits ofl on X are Xi = {Xij : j < n} for i < m and those of
g are Yj = {Xij : i < m} for j < n, and the action of], g on X is entirely determined
from that on] on Xo and ofg on Yo by Xij/kg/ = Xi'j' ¢} XOj/k = XOj' & Xiog/ = Xi'O
(so that ], g essentially act as a direct product). Or, for another example, we may
76 J. K. Truss

have IXI = 60 and each of],g acting regularly on X; this possibility arises in view
of the fact that the left and right regular actions of A(5) on itself commute.
The key to devising a suitable way of expressing disjointness is to look at those
faithful actions which are indecomposable, in the sense that they cannot be non-
trivially expressed as ] . 9 where ], 9 are two transitive actions. Now it is rather
clear that the faithful actions of degrees 60 and 30 are not indecomposable. This is
because if we take m = 12, n = 5 and m = 6, n = 5 respectively, allow], 9 to act
as a direct product, and consider] . g, we obtain faithful actions of degrees 60,30
(since m and n are coprime). It is equally clear on arithmetical grounds that the
faithful transitive actions of smaller degree are indecomposable, but this is not good
enough. We have to consider faithful, not necessarily transitive actions, where the
orbits are of these lengths, and see whether they can be built up as ] . 9 where ], 9
are commuting faithful actions.
Having sketched the ideas behind what follows, I state the technical lemmas we
need (proved in [19]).

Lemma 4.3 Suppose that Hand f{ are proper subgroups of A(5). Then for some
a E A(5), IH n a-I Kal :=:; 3. Moreover, if there is a such that IH n a-I J{al = 3 but
no b such that IH n b- 1 Kbl < 3, then IHI = IKI = 12.

The purpose of this lemma is to show that in product-like actions derived as in

the above sketch from non-trivial] and g,] . 9 has orbits of length at least 20.

Lemma 4.4 Let D be the diagonal subgroup {(ai, ai) : 1 :=:; i :=:; 60} of A(5) X A(5).
Then for any subgroup H of A(5) x A(5) of order 12 or 36, there is a such that
la-1HanDI i- 3.

This lemma further analyses the final clause of Lemma 4.3. Note that the
diagonal subgroup corresponds to the product] . 9 of the two associated coset
actions of A(5).

Lemma 4.5 Suppose that ],g are subgroups of Sym(X) isomorphic to A(5) (in
the specified listings) which centralize each other, and such that (f, g) is transitive
on X. Then]· 9 has an orbit of length at least 20. Moreover, if]· 9 has an orbit
of length 20 then it also has an orbit of some other length greater than 1.

This lemma reveals explicitly the purpose of the two previous lemmas. Its proof
involves the consideration of three cases: the first where ], 9 each act transitively
on X, in which case their actions will be regular; the second where one action is
transitive and the other is not, which turns out to be impossible; and the. third
where neither action on its own is transitive, and this is where Lemmas 4.3 and 4.4
are used.
We now move towards the construction of a formula which is intended to say
that x acts as A(5) on all but a small set of its orbits, and that each such orbit has
Recovering Structures from Quotients of their Automorphism Groups 77

length 1 or 5. Actually we stop short of doing this (even though it can be done) and
just find a formula restricting the range of representations possible, as this provides
a quicker route to our goal. We require the following auxiliary formulae:

comm(x,1/): A.l5.i<m,l5.j5. n XiYj =

YjXi,
where m and n are the lengths of x and y. This asserts that
each entry of x commutes with each entry of y.
conj(x, 1/): (3z)(x Z = 1/). _
indec(x): diag(A(5), x~V1/)(\fZ)(comm(Y, z)" diag(A(5), Ii) "
diag(A(5) , z)" x =
y. z ---> (conj(x, 1/) V conj(x, z)).
Lemma 4.6 For any sequence 1 of elements of S).. (/1) of length 60,
S).. (/1)/ S" (/1) F= indec( S" (/1).f) if and only if (1) acts as A(5) on all orbits outside a
small subset of /1, lI30(1) , lI60(1) < K, and there is at most one mE {5, 6,10,12,15, 20}
for which lIm (1) ~ K.

Proof Let us say that 1 ¢ S,,(/1) is indecomposable if

1
If lim (1) ~ K for two values of m > 1 then may be expressed in the form "9 . h
where "9 is the restriction of 1 to the union of its orbits of length m for one such value
of m, and h is its restriction to the complement. Since S,,(/1)."9 and S,,(/1).h clearly
cannot be conjugate to S,,(/1).f, this violates indec(S" (/1).f). If 1/30(1) ~ K then we
may express 1 as a commuting product "9. h of "9, h for which lI6("9) ~ K, 1/5(h) ~ K,
contrary to indec, and similarly if 1/60(1) ~ K, with 12,5 instead of 6, 5.
Conversely if m E {5, 6,10,12,15, 20} is the unique value for which lIm (1) ~ K,
the fact that indec(S,,(/1).f) holds is verified by appeal to Lemma 4.5.

We are now able to express 'disjointness', which (as we saw in studying A(O)
in §3) is the key to recovering the appropriate ring of sets. We shall refine this in a
moment so that we can apply it to involutions, and then to arbitrary members of
S)..(/1)/S" (/1). But for now we use elements satisfying indec. Let us say that two
indecomposable elements S,,(/1).f and 8,,(/1)."9 have the same action if lIm (1) ~ K
and I/mCg) ~ K for some m > l.

disJt (x, 1/) : indec(x) " indec(1/) " comm(x, y) " indec(x . 1/).

Lemma 4.7 For any sequences 1 and "9 of elements of S>.(/1) - S,,(/1) of length
60, S>.(/1)/S" (/1) F= disJt(8" (/1)1, S,,(/1)"9) if and only if 1 and "9 are indecomposable
with the same action, and Isupp 1 n supp "91 < K.

Proof It is clear that if two indecomposable elements of 8>.(/1)/8,,(/1) have

the same action, and almost disjoint supports (meaning that the intersection of
their supports has cardinality less than K), then they commute, and their point-
wise product is also indecomposable. Conversely suppose that 8>. (/1)/8" (/1) F=
78 J. K. Truss

disjl(S" (p)], S,,(p)g). Then as in the previous proof, if the supports of f and g
are not almost disjoint, then indecomposability of] . g is violated. It also follows
that ], g, and] . g must all have the same action.

We actually employ involutions to encode subsets of p. The use of A(5) en route

was for technical reasons, but since A(5) has elements of order 2 the fact that we
can express disjointness of indecomposables having the same action means that we
can also express disjointness of involutions. In fact it is easy to check that in its
transitive action of degree 12, A(5) has fixed-point free elements of order 2. Let
aj be one such. Then for 9 E S>.(p), S,.(p).g has order 2 if and only if there is
some indecomposable g with 1I12(g) 2: /'C such that S,,(p).g = S,,(p).gj. (If aj had
fixed points and A = p+ then we might have to 'piece together' such a 9 from two
moieties, which would only slightly complicate the argument).

dish(x, y) : x 2 = y2 = 1/\ 3z3t(zj = x /\ tj = Y /\ dish(z, t».

disj(x, y) : 3xax 23 y l 3Y2(X = Xl X2 /\ Y = Yl Y2 /\ Al5,i,j 5,2 dish(x;, Yj»·

Lemma 4.8 (i) For any f, 9 E S>.(p), S>.(p)/ S,,(p) F dish(S" (p).f, S,,(p).g) if
and only if S" (p).f and S" (p ).g are involutions (or the identity) such that Isupp(f)n
supp(g) I < /'C.
(ii) For any f,g E S>.(p),S>.(p)/S,,(p) F disj(S,,(p)·f,S,,(p).g) if and only if
Isupp(f) n supp(g) I < /'C

Proof (ii) follows from the fact that any permutation may be written as a
product of two involutions.

In terms of 'disjointness' we can now move quickly to express the other properties
required to interpret the suitable (quotient) ring of sets in
S>.(p)/S,,(p). We state the the following two formulae and what they express with-
out proof.

subset(x, y) : Vz(disj(y, z) ---. disj(x, z».

restr(x, y) : 3z(disj(x, z) /\ xz = y).
Then S>.(p)/S,,(p) F subset(S" (J-l).f, S,,(p).g) {:} Isupp(f) - supp(g)1 < /'C, and
S>.(J-l)/S,,(J-l) F resir(S,.(J-l).f, S,,(J-l).g) {:} for some f' E S,,(J-l)·f, g' E S,,(J-l).g, f' is
a restriction of g'.
The following theorem is (essentially) the same as Theorem 4.2 of [11], though
Rubin's methods are somewhat different (and in particular do not make use of the
'trick' involving A(5».

Theorem 4.9 Let P>. (J-l) , P,,(J-l) be the rings of subsets of p of cardinality less than
A,/'C respectively (in the language {=,U,n,~}), ThenP>.(p)/P,,(p) is interpretable
in S>.(p)/S,,(p).
 Recovering Structures from Quotients of their Automorphism Groups 79

Proof. Let D be the family of involutions in S).. (ft)/ S,,(ft) and for S" (ft).g E Diet
F(S,,(ft).g) be the equivalence class in P)..(ft)/P,,(ft) of supp(g). Then F(S,,(ft).!) =
F(S,.(ft)·g) {:> S)..(ft)/S,.(ft) F= subset(S,.(ft).f, S,,(ft)·g) /\ subset(S" (ft).g, S,,(ft).f),
and 'subset' also enables the boolean operations (except for complementation) to
be expressed. For instance unions may be expressed by the formula (needed later)
union(x, y, z) : Vt(subset(x, t) /\ subset(y, t) +-+ subset(z, t».
Now we move on to attempt to get information about outer automorphisms. In
particular we show how the methods presented here can be used to prove the result
of [1]. Unfortunately, unlike in §§2,3, the interpretability of P)..(ft)/P,,(ft) does not
appear to be sufficient to derive this without extra work, and the same applies in
§5. What is required rather is to be able to interpret the family of sequences of a
suitable length (as ever modulo less than", 'mistakes') inside s).. (ft)/S" (ft)· To do
this we use very similar ideas to [1]; the main difference is that the reasoning there
is mostly second order (a study of centralizers being one of the key points). The
methods at present only work for", = w, which is assumed from now on, and I list
various formulae which are needed.

minimal(x) : x f:. 1/\ Vy(restr(y, x) - t (y = 1 Vy = x».

fincycle(x) : Vy(restr(y, x) -+ -,minimal(y» /\ 3y(y f:. 1/\ disj(x, y».
norepeat(x) : fincycle( x) /\ VyV z(y f:. 1/\ disj(y, yZ) -+ -,restr(yyZ , x».
allfincycle(x) : norepeat(x) /\ Vy3z(norepeat(y) -+ restr(yZ, x».
infcycle(x) : Vy(restr(y, x) /\ y f:. 1 -+ -,fincycle(y».
cut( x, y) : Vz( z f:. 1/\ restr( z, x) -+ (-,disj( z, y) /\ ...,subset( z, y»/\
Vz(z f:. 1/\ restr(z, y) -+ (-,disj(z, x) /\ ...,subset(z, x))).
seq(x, y) : x f:. 1/\ x 2 = 1/\ allfincycle(y) /\ cut(x, y) /\ 3z(z f:. 1/\
disj(x, z) /\ disj(y, z».
sameseql (Xl, Yl, X2, Y2) : seq(xl' yt} /\ seq(x2' Y2) /\ 3z(yf = Y2 /\ disj( Xl, z)/\
disj(X2' z».
sameseq(xl, Yl, X2, Y2) : 3xi 3Yi 3x~3Y2(sameseql (Xl, Yl, xi, yD/\
sameseql (xi, yi, x~, y~) /\ sameseql (x~, y~, X2, Y2».
The intended meanings of these should be fairly clear, and I sum up their prop-
erties in the next lemma.

Lemma 4.10 (i) S)..(J-l)/Sw(J-l) F= minimal(x) if and only if there is g having a

single infinite cycle and fixing all other points such that x = Sw(J-l).g.
(ii) S>.(J-l)/Sw(J-l) F= fincycle(Sw(J-l).g) if and only if all cycles of g are finite
and g has infinitely many fixed points.
(iii) S)..(J-l)/Sw(J-l) F= norepeat(x) if and only if there is g having all cycles finite,
infinitely many fixed points, and no two cycles of the same length greater than 1
such that x = Sw(J-l).g.
(iv) S>.(J-l)/Sw(J-l) F= allfincycle(x) if and only if there is g as in (iii) having a
unique cycle of every possible finite length greater than 1.
80 J. K. Truss

(v) S>.(/-l)/Sw(/-l) F= inJcycle(x) if and only if for some 9 with Sw(/-l).g = x,g
has no non-trivial finite cycles.
(vi) S>.(/-l)/Sw(/-l) F= cut(x, y) if and only if there are g, h with Sw(/-l).g = x and
Sw (/-l).h = y such that every non-trivial cycle of 9 intersects some non-trivial cycle
of h and contains a fixed point of h, and vice versa.
(vii) S>.(/-l)/ Sw(/-l) F= seq(x, y) if and only if there are 9 of order 2 and h all
of whose cycles are finite and having a unique cycle of each length greater than
I, with Sw(/-l).g = x and Sw(/-l).h = y and /-l - (supp(g) U supp(h» infinite, and
the non-trivial cycles of 9 and h can be put into 1 - 1 correspondence 'P by letting
G'" = G' {:;> IG n G'I = l.
(viii) S>. (/-l)/ Sw (/-l) F= sameseq( Xl, Yl, X2, Y2) if and only if there are gl, hI, g2, h2
as for seq such that for each n > 1 the intersections of the cycles of hI and h2 of
length n with the supports of gl, g2 respectively, are equal.

The idea here is that we are now able to encode (countable) sequences of mem-
bers of /-l by pairs (g,h) where S>.(/-l)/Sw(/-l) F= seq(Sw(/-l).g,Sw(I-l).h). For n > 1
the nth entry of the sequence so encoded is to be the unique member of the cycle of
h of length n which intersects supp(g). Since this appears to depend on the choices
of 9 and h, we have to be able to express the fact that two choices encode the same
sequence, and this is what sameseq is for.
In order to indicate what is now required, let us suppose that Xl and X 2 are
sets and that () is an isomorphism from S>.(Xd/Sw(Xd to S>.(X 2)/Sw(X2 ). The
aim is to show that this is induced by a bijection J from a cofinite subset of Xl to a
cofinite subset of X 2 . For any countable sequence a:::: (ao, aI, ... ) of members of Xl
we find a partial 1 - 1 map Pa from its set of entries into X 2, via the interpretation
of sequences in S>.(Xd/Sw(Xd as given in the previous paragraph. Let (g, h) be
chosen as described, and let Sw(X 2).g', Sw(X 2).h' be the images of Sw(Xr).g and
Sw(Xd.h under (). Since S>.(Xd/Sw(Xd F= seq(Sw(Xr).g,Sw(Xr).h), and () is an
isomorphism, S>.(X 2 )/Sw(X2 ) F= seq(Sw(X 2).g', Sw(X 2).h'). By Lemma 4.10(vii),
there is a sequence a' encoded by (Sw(X 2 ).g', Sw(X 2 ).h'), by Lemma 4.10(viii) this
depends only on a (up to finitely many mistakes) and not on 9 and h, and we let
Pa be given by aiPa = a:.
The main extra piece of work which is needed is to show that these 'approx-
imations' to the map J are compatible. Here what is required is that if b is a
subsequence of a, then {i : biPa f. biPb} is finite. For this we devise a method of
'storing' the behaviour of a subsequence of a sequence, and the following formulae
are introduced with this in mind.

store n (Xl, ... , Xn , Yl, ... , Yn, z) : inJcycle(z) /\ /\l~i,j~n disj(Xi, Yj)/\
/\l<i<n((XiYiY = Yi) /\ Vt(restr(t,z) /\ t f. z ---t Vl<i<ndisj(t,Xi»,
- - (;;;; 1,2,3, ... ).
equaIJincycle(x) : Jincycle(x)/\VyVz(disj(y, z)/\x = yz ---+ (conj(x, y)Vconj(x, z».
cyclesn(x): equaIJincycle(x) /\ xn = 1 /\ /\l~m<n(Xm f 1), (n = 2,3, ... ).
 Recovering Structures from Quotients of their Automorphism Groups 81

init;j(x, y, z) : cycles2(x) /\ cycles;(y) /\ cyclesj (z) /\ cut(x, y) /\ cut(x, z)/\

cyclesU_i+l)(y-1z), (2 ~ i < j).

Lemma 4.11 (i) S>..(p)/Sw(p) F= storen (x1, ... , X n , Yl,"" Yn, z) if and only if
there are gi, hi, k such that each gi and hj have disjoint supports, the non-trivial
cycles of k are all infinite, U1 <i<n SUpp(9i) forms a transversal for these infinite cy-
cles of k, each hi is the product afthe conjugates of gi by positive powers of k (where
the product of an infinite sequence of pairwise disjoint permutations is defined in
the natural way), and Sw(p).gi = Xi, Sw(l1,).hi = Vi, Sw(p).k = z.
(ii) S>..(p)/Sw(J-t) F= equalfincycle(x) if and only if there is g having all non-
trivial cycles of the same finite length such that x = Sw(J-t).g.
(iii) S>..(J-t)/Sw(p) F= cyclesn(x) if and only if there is 9 having all non-trivial
cycles of length n such that X = Sw(J-t).g.
(iv) S>..(J-t)/Sw(J-t) F= init;j(x,y,z) if and only if there are g,h,k such that X =
Sw(p).g, y = Sw(J-t).h, z = Sw(p).k, all non-trivial cycles of g, h, k have lengths
2, i, j respectively, the cycles of 9 and h are in 1 - 1 correspondence 'P given by
G' = G'P {::} IG n G'I = 1, and similarly for the cycles of 9 and k, and if (0:0 13),
(0:00:1 ... O:i-d, and (0:0 o:~ ... 0:}_1) are cycles of g, h, k intersecting in {o:o} then
(0:0,0:1, ... , O:i- d is an initial segment of (0:0, o:~, ... , o:} _ d·

Lemma 4.12 Let a be a countable sequence of members of Xl, and let h be an

infinite subsequence. Let Pa, Ph be the 1-1 maps from {ai : i E w} and {b j : jEw}
into X 2 respectively, determined as above (by choice of suitable pairs satisfying seq).
Then {j : bj Pa i- bj Ph} is finite.

Proof Find g, h as in the definition of Pa with Xl - (supp(g) u supp(h»

infinite, to allow 'spare space for storage'. Let us write 9 = TI~2(ai Ci) and
h = TI~2(ai ail ai2·· .ai i-d where all the ai,Ci,aij are distinct. Let

B = {i : 3j(bj = ai)}

and define gl and h1 by gl = TIiEB,i>2(ai Ci)' hI = fl EB,i>2(ai ail ai2··· ai j.-I)

where for i E B, ai = bj ,. Then (gl,-hd is suitable for the definition of Ph' We
have to 'copy' (g, h, gl, ht) infinitely often to ensure that its pattern is not lost even
under the distortion caused by passing to the quotient by Sw(Xt).
Find kl,k2,k3,k4,1 which 'store' (g,h,gl,ht). For this we first choose I having
just infinite cycles (and fixed points) and so that each infinite cycle of I contains
just one member of supp(g) U supp(h). Then we let ak l = C {::} (3n > O)(al-ng =
cl- n ), ak2 = C {::} (3n > O)(al-nh = cl- n ), and similarly for k3 and k 4. Thus
S>..(Xl )/ Sw(Xd F= store(Sw(Xt).g, Sw(Xt}.h, Sw(Xt}.gl,
Sw(Xt).h l , Sw(Xl ).k l , Sw(Xl ).k 2 , Sw(X l ).k3, Sw(X 1 ).k 4, Sw(X l ).1). In fact we could
say that (k l ,k2 ,k3,k4,1) 'stores' (g,h,gl,hd 'accurately' even in S>..(Xt). The
images of the nine cosets under () also satisfy store in S>..(X2)/S,,-,(X 2 ), and this
82 J. K. Truss

means that there must be gl,hl,g~,h~,k~,k~,k~,k~,l' such that (Sw(XI).g)B =

Sw(X 2).gl and so 011, and where for suitable choices of distinct a;, C;, a;j E X 2, gl =
Il~2(a; cD, hi = Il~2(a; a;1 a;2" .a; i-I) and k;,k;, [I are determined fromg l and
hi as k 1 , k z , [were from 9 and h.
We show that. (up to finitely many mistakes) g~ = IliEB i>Z(a; bi) and hi =
a:
IliEB,i>z(a: 1 a;2'" a; j i - 1 )' Let i E B. Then if 1"1, rz, 1"4 a~e-the restrictions of
k 1, k2' £4 to the union of the cycles of I intersecting {ai, Ci, aiO,
... , ai j,-d, then S;>..(p)/Sw(p) 1= l·eslr(Sw(p).I'I, Sw(Jl.).kJ) 1\ restr(Sw(p,).rz,
Sw(,l).k 2 ) 1\ restr(Sw(p).r4' Sw(P)·k4) 1\ init ji i(Sw(p).k 1 , Sw(P).k4' Sw(p).k2)' This
formula is preserved under e, and it follows that for all but finitely many cycles
of r~, 7'S, r~ (representatives of the corresponding images), the non-trivial cycles of
1';, r~ can he put into 1 - 1 correspondence with those of r~ by virtue of having 1l01l-
empty intersection, they have lengths 2, ji, i, and they are restrictions of k~, k;, k~
respectively. It follows that IliEB,i>2(a: cD and IliEB,i>2(a; a;1 a;2'" a; j,-l) serve
for g~ and h~ as in the definition 'Of Pb, and so must -actually be equal to g~ and
h~ apart from finitely many mist.akes. Therefore {i E B : bjiPb # aD is finite, and
since {i : aiPa # aD is finite, we deduce that {j : bj Pb # bj Pa} is also finite, as
required.

Theorem 4.13 Let Xl and X 2 be countable sets, and let 0 be an isomorphzsm

from S(XIl/Sw(Xd to S(X 2 )/Sw(X2 ). Then there are finite sets A ~ Xl and
B ~ X 2 and a bijection f : (Xl - A) (X 2 - B) such that for every 9 E
---4

S(XIl, (Sw(Xd·g)B = S,,-,(Xz).f-lgf.

Proof. We relllark that stridly speaking, J-l g f does not lie in S(Xz), which
means that we cannot form the coset. Sw(X 2 ).f- l gf. But f-l gf maps X 2 - (B U
(Ag- 1 - A)!) 1 - 1 onto X 2 - (B U (Ag - A)!), and as IB U (.4g- 1 - A)fl =
I(Bu(Ag-A)fl, there is liE S(X 2 ) extending f- 1 gf, and so we may unambiguously
define Sw(X 2 ).f- 1 gf to equal Sw(X 2 ).h for any such h.
Let Xl = {Oi: iEw}. Then ifao = (00,a2,04, ... ) and al =
(ai, 03, 05, ... ), Pao and Pal arc both defined, and we let f' = Pao U Pal' We show
that f' is a 1-1 map from Xl into X 2, apart from finitely many exceptions, meaning
as in the statement of the theorem that there arc finite subsets A ~ Xl and B ~ X 2
such that if f is the restrictioll of f' to Xl - A, then J is a bijection from Xl - A
to X 2 - B (and also has the property required, in t.hat 'conjugation' by f induces
e 011 the quotients).

First let Y be the intersection of the ranges of Pao and Pal' Let a l be a sequence
having ao as a S11 bsequence, and having the members of Y P~; as its other entries. By
Lemma 4.12, C = {f12i : (J.2iPao # (!2iPa'} and D = {a2j+l : 0:!j+1Pa l # (12j+1Pa'}
arc finite. VVe show that }' is conta.ined in CPa o U Dpa, , and hence is finite. If not
t.here is (!2iPa o = (!2j + 1 Pal E Y - (CPa o U Dpa, ). Thus (J2i tI. C and (J.2j + I fi D and
so 02iPa' = 02i]Ja o = (/2j +lPa l = (/2j +l]Ja', cont.rary to Pal 1 - 1.
Recovering Structures from Quotients of their Automorphism Groups 83

We deduce that Pao U Pa, is 1 - 1 on Xl - A where A = {a2i : 3j(a2iPao =

a2j+lP a ,)}, Let f = f'1(X1 - A). Since the whole argument may evidently be
reversed, we deduce also that B = X 2 - (Xl - A)f is finite.
Finally we have to show that for each 9 E S(Xd, (Sw(Xd·g)8 =
Sw(X 2 ).f- 1gf (bearing in mind the above-mentioned proviso). Let a be a fixed 1-1
sequence with coinfinite range, and let (gl, h 1 ) be a pair representing a as in the def-
inition of Pa. Then ((Sw(X 1).gl)8, (Sw(Xd·hd8) represents af = a'. Moreover, ag
is represented by (gf, hi) and so (ag)l is represented by (Sw(XI).gf, Sw(XI).hf)B =
((Sw(XI).gi)8, (Sw(XI).hf)B) = ((Sw(xI)gd, (Sw(XI).hI)8)(Sw(x').d. But this
also represents (af)(Sw(x,).g)8, and so we deduce that up to finitely many mistakes
(ag)f = (af)g', where Sw(X 2 ).g' = (Sw(XI).g)8. We may now let a equal ao
and a1 in turn, and deduce that {i : a;gf "I a;Jg'} is finite, and therefore that
Sw(XI).gf = Sw(X 1)·fg', from which (Sw(XI).g)8 = Sw(X 2).f- 1gf follows.

Corollary 4.14 [1] The outer automorphism group ofS(w)/Sw(w) is infinite cyclic,
and is generated by () given by l = =
f-1 gf where nf n + 1 for all n.

Now let us remark on possible extensions of these results to other values of J-l.
At present the restriction to K, = W seems unavoidable, without use of stronger
methods. Subject to this the general problem then boils down to showing that the
various 'approximations' Pa can be consistently fitted together to make one big map
f which agrees with each of them on all but finitely many points. The particular
case analysed in Theorem 4.13 worked because the whole of the set could be covered
by two countable coinfinite sets. The best we can do at present uses a similar idea,
but with>' = J-l+ (in other words where the top group is the full symmetric group).

Theorem 4.15 Let Xl and X 2 be sets of cardinality J-l, and let () be an isomor-
phism from S(Xd/Sw(XI) to S(X 2 )/Sw(X 2). Then there are finite sets A ~ Xl
and B ~ X 2 and a bijection f : (Xl - A) --+ (X2 - B) such that for every
9 E S(XI) , (Sw(XI).g)B = Sw(X 2 )·f-1 gf· Therefore the outer automorphism group
of S(J-l)/Sw(J-l) is infinite cyclic and is generated by () given by l = f-1 gf where
nf = n + 1 for all nEw, and oJ = cy for infinite CY < J-l.
Proof. We just indicate the extra trick required. Instead of using single countable
sequences a, we use sets A = {a", : CY E J-l} of pairwise disjoint countable sequences
of elements of J-l, the complement of whose union has cardinality J-l (so that the
range is a 'moiety'). We require modifications of the formulae previously used to
encapsulate this new situation. In place of seq we shall have a formula setof seq
with three variables. The third will satisfy infcycle and each of its (infinite) cycles
will be the union of those cycles of the first two elements which encode a", for a
fixed CY. There will also be a formula samesetofseq saying that two of these triples
encode the same sets of sequences.
Given that this can be expressed, we are able to define a 1 - 1 map PA for any
such set A of sequences which takes the union of the ranges of its elements into X 2 .
84 J. K. Truss

Moreover, for any countable subset of A, the situation is as before, since everything
can be encoded into just one sequence. This means for instance that we do not need
to store the whole of such a set (though we could). For the analogue of Lemma 4.12
actually follows from that case. This is because if {b : bPB =/; bpA} is infinite, then
this fact may be established by looking at just a countable subset of A, which will
then contradict Lemma 4.12.
The proof of the theorem is completed as for Theorem 4.13 using the fact that
Xl may be covered by two moieties.

5 Quotients of Automorphism Groups of Chains

In this section I carry out a similar exercise to the one described for quotients of
symmetric groups in §4, but here for quotients of automorphism groups of chains.
This work was described in [5], and so I shall just highlight the most significant
points, and derive the modified statement (about recovery of the 'tail' of a struc-
ttue) , rather than concentrating on the outer automorphism problem. Moreover,
although a certain amount of generality was achieved in [5] at some points, the main
goal was to show that A(R)j L(R) has no outer automorphisms, and so T shall focus
on that case here. This has the effect of simplifying some of the lemmas.
The notation for automorphism groups of chains Q was all introduced in §3, so
we can begin by listing the key properties of A(Q)j L(Q) which we intend to express
in the first order language of groups. One major difference from §3 is that this
time the (lattice-) ordering on A(R)j L(R) is definable from the abstract group.
The 'trick' is reminiscent of the one used in formulating the property camp (in §3).
Another point is that for the case of A(0) we were able to capture points of IT by
usc of the formula half. In the quotient group we instead have to aim to interpret
unbounded subsets or sequences of Q (or IT). If as in §4 we try to use countable
sequences then these will only exist in the case of countable cofinality, and things
work most easily for R (and Q), since in addition all points (even of the completion)
are limits of countable increasing or decreasing sequences.
In view of this, we shall use the formulae conjn mentioned earlier, which (for
n = 1,2) work well in this case.
conjI(x) : x=/; 1/\ Vy3z(xxY = .TZ).
conj2(x) : .r #- 1 /\ ,conj I( x) /\ 3u3vVy3z( xx Y = (xx" y V xxY = (n:V)2).

LeIllma 5.1 For n a doubly homogeneous chain havinq countable cofinality,

A(0)j L(0) F= conjl( r,(0).y) zf and only If g has an unbounded orbital.

Proof. If g has an unbounded orbital, then so does any conjugate, and these
orbitals have the same parity (±I). The product also then has an unbounded
orbital which is again of the same parity (and which contains the intersection of
the individual orbitals given), and so, by applying Holland's criterion (mod L(Q))
is again conj ugate t.o g.
Recovering Structures from Quotients of their Automorphism Groups 85

Conversely, if 9 does not have an unbounded orbital, then as it is not 1 (in the
quotient group), it has a cofinal collection of non-trivial orbitals (an, bn ) say with
an < bn ::; an+l for each n. Moreover by passing to an infinite subset we may
suppose that these all have the same parity. By taking the product of L(O).g with
a suitable conjugate (whose corresponding orbitals contain [b n , anH]) we obtain
an element which does have an unbounded orbital, and which cannot therefore be
conjugate to L(O).g.
It is shown in [5] that there are (at least) two possible kinds of elements in
A(O)/ L(O) which may satisfy conj2, called there 'contiguous' and 'separated'. If
o is order-complete then no elements can be separated, but if 0 is not order-
complete, then both possibilities arise, and how many orbits they fall into depends
on the number of orbits of A(O) on n. We say that 9 is contiguous if there is an
increasing divergent sequence (ao, aI, ... ) in n, all entries of which lie in the same
A(O)-orbit, such that (beyond some point) all (an, an+l) are non-trivial orbitals of
9 of the same parity. It is said to be separated if there is such a sequence in nsuch
that all (a2n, a2n+d are non-trivial orbitals of the same parity, 9 fixes all points of
each (a2n+l, a2n+2) pointwise, and there are distinct orbits X and Y of the action
of A(O) on nsuch that each a2n lies in X and each a2n+l lies in Y.

Lemma 5.2 For any 9 E A(R), A(R)/ L(R) F= conj2(L(R).g) if and only if 9 is
contiguous.
Proof. If 9 is contiguous, then the proof that A(R)/ L(R) F= conj2(L(R).g)
is rather similar to that of the previous lemma. On taking the product of 9 with
a conjugate, either all but finitely many of the fixed points are deleted, in which
case the resulting element has an unbounded orbital, of parity the same as the
eventual common parity of the non-trivial orbitals of g, or infinitely many fixed
points remain, in which case the resulting product is still conjugate (mod L(O)) to
g. (The 'eventual' fixed point set of the product of 9 with its conjugate is just the
intersection of the two sets of fixed points). Thus two conjugacy classes result.
Conversely suppose that A(R)/ L(R) F= conj2(L(R).g). Since
-'conj1(L(R).g), 9 does not have an unbounded orbital, and since 9 (j. L(R), it has
a cofinal set of non-trivial orbitals ordered in type w. By passing to an infinite
subset we may suppose that these all have the same parity, +1 say. By taking
products of 9 with suitable conjugates, we can build (a) a contiguous element, and
(b) an element with an unbounded orbital. Since we are assuming that there are
only two such elements in A(R)/ L(R) up to conjugacy, it follows that 9 itself must
be contiguous.
In terms of these special elements we now give the formulae used to recapture
the lattice ordering.
pos1(x) : conj1(x) 1\ VyVz(conj2(y) 1\ (3u)yyZ = yU --+ (3u)yyZ = y",u).
pos(x) : (3y)(pos1(xxY)) 1\ (Vy)-,pos1(x-1(x-1)y).
86 J. K. Truss

Lemma 5.3 For any g E A(R), A(R)/ L(R) ~ pos1(L(R).g) if and only if g has
an unbounded orbital of parity + 1.

Proof Choose a contiguous element h having positive orbitals (n, n + 1) for

nEw. Then we know that for any k, L(R).hh k either has an unbounded orbital,
or is conjugate to L(R).h. In the latter case, a conjugator must take (a tail of)
w to (a tail of) the fixed point set of hhk. Now this fixed point set is a subset of
w, so if L(R).g has an unbounded orbital of parity +1, a conjugator of L(R).h to
L(R).hh k can be found which is itself conjugate to L(R).g. If on the other hand,
for any such k a suitable conjugate of L(R).g exists, then we can choose k so that
the fixed point set of hhk contains eventually just the even integers, and this means
that the conjugator, and hence also L(R).g, cannot be negative, and as g has an
unbounded orbital, it must have parity +l.
Let us say that L(R).g is positive if {a : a < ag} is unbounded above, but
{a : ag < a} is not. This notion of positivity corresponds to the natural ordering
on A(R)/ L(R) induced from the pointwise ordering on A(R).

Lemma 5.4 For any g E A(R), A(R)/ L(R) ~ pos(L(R).g) if and only if L(R).g
is positive.

Proof If {a : a < ag} is unbounded above, then there is h such that ggh has an
unbounded orbital of parity + 1, and if {a : ag < a} is unbounded above, there is h
such that ggh has an unbounded orbital of parity -1. The given condition therefore
precisely says that {a : a < ag} is unbounded above, but {a: ag < a} is not.
Once one has shown how to define the positive elements inside the group, it
is then easy to define the (lattice-) ordering: g ~ h {::} gh- 1 ~ 1, and from this
many other things can be defined, of course including the lattice operations 1\
and V, 'absolute value' Igl = g V g-l, and also in terms of these, disjointness of
supports, containment, and restriction. All these will apply eventually; for instance,
saying that L(R).g and L(R).h have 'disjoint' supports here is taken to mean that
supp(g) n supp(h) is bounded above. In addition to this proviso, saying that the
support of L(R).g is contained in that of L(R).h means that supp(g) - supp(h) is
bounded above (compare the proof of Theorem 3.8).

disj(x, y) : Ixl/\ Iyl = 1.

subset(x, y) : (Vz)(disj(y, z) ---+ disj(x, z».
restr(x, y) : (:lz)(disj(x, z) /\ xz = y).
The main technical point needed to prove that A(R)/ L(R) has no outer auto-
morphisms is to show how to interpret the family of strictly increasing divergent
sequences a = (an: nEw) of reals in A(R)/L(R) (up to finitely many mistakes).
There are various ways to do this, but the one chosen, as for the case of S).. (/J)/ Sw (/J),
employs a pair of elements, the first to serve as a marker (or rather to provide a
Recovering Structures from Quotients of their Automorphism Groups 87

sequence of markers), and the other to encode the increasing sequence of positive
integers. To build up to this, we first need a formula distinguishing those positive
elements whose support (eventually) has the form U{(a2n,a2n+d : nEw} (which
are the first kind of element).
moil(x) : pos(x) /I. ~y(y :/; 1/1. disj(x, y) /I. conj2(xy).
moi(x) : moil(x) /I. (Vy)(~z)(moil(y) --> restr(x Z , y).
By a moiety we understand a positive element 9 of A(R) whose non-trivial
orbitals are of the form (a2n, a2n+l) for some strictly increasing divergent sequence
a of reals.
Lemma 5.5 For any 9 E A(R), A(R)j L(R) ~ moi(x) if and only if x = L(R).g
for some moiety g.
Proof Let (an : nEw) be a typical strictly increasing divergent sequence. Its
entries thus form the (eventual) fixed point set of an element satisfying conj2. The
formula moil thus says that an infinite coinfinite subset of the set of the intervals
(an' an+l) is chosen as the (eventual) family of non-trivial (and positive) orbitals
of the element in question. The extra condition imposed by moi is just that the
element should be 'minimal' among such elements, or more precisely, it can be
conjugated to a suitable restriction of any element satisfying moil. This has the
effect of ensuring that for all but finitely many n, (an, an+d and (an+l' an+2) are
not both selected, so by relabelling the ans, the support can be expressed in the
desired form.
To encode strictly increasing sequences (an : nEw) we use pairs (g, h) of
moieties such that the non-trivial orbitals of 9 have the form (an, bn ) for some
bn E (an, an+d, and for each n there are exactly n non-trivial orbitals of h contained
in (b n , an+d.
link(x, y) : disj(x, y) /I. moi(xy) /I. x, y:/; 1/1. Vz[restr(z, x) /I. ~u~v(disj(u, y)/I.
disj(v, y) /I. z = XUx") --> z = xl.
inc(x, y) : link(x, y) /I. ~dt[disj(z, t) /I. Y = zt /l.link(t, x) /l.link(x, t)/I.
~u(pos( u) /I. disj( u, z) /I. ttl = x) /I. ~u~v~w(pos( u) /I. pose v) /I. pose w)/I.
disj(u,x)/l.disj(v,t)/l.disj(w,yU)/l.yllV = z/l.xW =tJ.
Lemma 5.6 (i) For any p and q, A(R)j L(R) ~ link(p, q) if and only if there are
moieties 9 and h in A(R) with disjoint supports and with no two positive orbitals of
9 or h Ii/taring an endpoint, such that p = L(R).g and q = L(R).h, and in between
any two positive orbitals of 9 there is at least one positive orbital of h.
(ii) For any p and q, A(R)j L(R) ~ inc(p, q) if and only if there are moieties
9 and h in A(R) with disjoint supports and with no two positive orbitals of 9 or h
sharing an endpoint, such that p = [,(R).g and q = L(R).h, and such that if the
positive orbitals of 9 are (an, bn ) for nEw WIth ao < bo < al < bi < a2 ... then for
each n there are exactly n + 1 positive orbitals of h contained in (b n , an+d.
88 J. K. Truss

Proof Suppose that 9 and h are as described in (i). Then gh is a moiety.

Let k be any element which can be expressed as a product of two conjugates by
conjugators disjoint from h. Then k must have at least one positive orbital in any
interval determined by the orbitals of h which contains a positive orbital of g, so if
it also is a restriction of g, must have exactly one. On passing to the quotient by
L(R), the conclusion is that k must agree with 9 on all but finitely many orbitals.
This verifies A(R)j L(R) 1= link(L(R).g, L(R).h).
Conversely if link holds for p and q then there are moieties 9 and h with p =
L(R).g and q = L(R).h having disjoint supports and with the positive orbitals of
g, h sharing no endpoints. Let (an, b,,) be the positive orbitals of g. Then it suffices
to show that for all but finitely many TI, there is at least one positive orbital of h in
(b n ,a l1+d. In cases where this is not true, if for instance (an,bn+d nsupp(h) = 0,
we can find witnesses k and / corresponding to u and v in link, with supports
disjoint from h, so that forming the product gk gl has the effect of 'coalescing' the
two orbitals (an, bn ) and (a n+l, b,,+d· By this method we may produce an element
which infinitely often has fewer positive orbitals in between consecutive positive
orbitals of h than 9 does, and 9 cannot then be a restriction of this element.
(ii) Rather than giving full details, I shall just explain the various parts of the
formula, which should indicate how a proof would go. If the given condition holds
for 9 and h, let (c n , dn ) be the last orbital of h in (b n , an+l) for each n, and take
as witnesses to z and t in inc the elements L(R).k and L(R).1 where I and k are
the restrictions of L(R).h to the union of all the (c n , d,,) and to the union of all
the other positive orbitals of h respectively. The fact that link(L(R).I, L(R).g) and
link(L(R).g, L(R).i) just corresponds to the alternation of the positive orbitals of 9
=
and i. The condition 3u(pos( u) I\disj( It, z) 1\ ttL x) asserts that the positive orbital
of i contained in any interval of positive orbitals of 9 is the last positive orbital of
h in this interval, which agrees with our choice. Since k has one fewer positive
orbital than II contained in each interval of positive orbitals of g, t.o conjugate
h to k (even modulo L(R)), a conjugator has to carry the positive orbitals of h
contained in (b n ,Cl ll +Il to the positive orbitals of k contained in (b n +1 ,a n +2) for
each n. The final clause in inc tells us that this can be done: It moves these orbitals
into (d n ,a n+1), and v moves them into (b n +1 ,a n +2); w iB needed to ensure that
after conjugating by tI, these orbitals have moved to the right of dn .
Conversely, by (i) we know that on the basis of link, if inc holds for p and q we
may find corresponding moieties g and h with disjoint supports such that if (an, bn )
for nEw are the positive orbitals of g, for each n there are mn positive orbitals of
h in (b", an+d where I :S mn < 0':>. Running the sketch of the previous paragraph
in reverse we find that for each n, m,,+l = ntn + 1, and this means that choices of
g and II can be made to meet the stated conditions.

Now that we can encode strictly increasing divergent sequences, we can move
towards the main result. For this we suppose that we are given an isomorphism
(j from A(fh)/ L(fh) to A(02)j L(02), where 0 1 and O 2 are doubly homogeneous
Recovering Structures from Quotients of their Automorphism Groups 89

chains, each having a tail isomorphic to that of R (so that in fact 0 1 , O 2 are iso-
morphic to either R or the 'left long line' L - , see [5]). The aim is to show that ()
is induced by some isomorphism between tails of 0 1 and O2 . We can now begin to
build up this desired isomorphism. We use the same notation inside each of 0 1 and
O 2 as we have been doing in R. In particular from now on we write a, b and so on
to stand for strictly increasing divergent sequences (ao, aI, a2, ... ), (bo, b1 , b2, ... ) in
0 1 or O 2 . For any such a in 0 1 choose moieties g, hE A(Od whose positive orbitals
share no endpoints, those of 9 are (an, bn ) for an < bn < an+l, and h has exactly
n positive orbitals in (b n , an +l) for each n. Since () carries the family of pairs in
A(Od/ L(Od satisfying inc to those in A(02)/ L(02) satisfying inc, by Lemma 5.6
there must be g' and h' with these same properties such that (L(01).g)9 = L(02).g'
and (L(Od.h)9 = L(02).h'. Let the positive orbitals of g' be (a~,b~). Then we
define Pa to have domain {an: nEw} and to take an to a~ for each n.
Of course we do not expect the whole of Pa to be accurately determined, but it is
meant to agree with the final isomorphism (which will carry. L(Od.g to (L(02).g)9
for all g) on all but finitely many points. To achieve this we need to verify the
following:
(i) the value of Pa is independent of the choice of 9 and h,
(ii) for any a and b,Pa and Pb are compatible,
(iii) there is some isomorphism k from a final segment ('tail') of 0 1 onto a final
segment of O2 whose restriction to the domain of any Pa agrees with it on all but
finitely many points,
(iv) if k is as given in (iii) (which is then clearly unique beyond some point), then
L(02).gk = (L(OI).g)o, (where gk, which is strictly speaking an isomorphism from
one final segment of O 2 to another, defines a unique coset L(02).gk = L(02).l where
l E A(02) is any extension of gk to an automorphism).
Lemma 5.7 Let (gl, hI) and (g2, h2) be two possible choices of (g, h) as used in
defining Pa, and let PI, P2 be the resulting partial maps. Then {n : an PI :f:. anP2} is
finite.
Proof We devise a formula sameseq analogous to the one used in §4.
between(x, y, z) : inc(x, y) A disj(xy, z) A moi(xyz) A 3u3v[disj(z, u)A
link(x, zu) A link(zu, x) A v> 1 A disj(v, y) A (zu)V = x].
sameseq(x, y, z, u) : inc(x, y) A inc(z, u) A 3v[subset(v, x) A 3w(disj(w, y)A
VW = x) A subset(v, z) A 3w(disj(w, u) A VW = z) A 3w(disj(v, w) A yW =
u)] A 'v'v[(subset(v, x) --- ...,between(z, u, v» A (subset(v, z) -+
...,between(x, y, v»].
These two formulae have the following intended meanings:
between(x, y, z): z has at most one positive orbital between any two consecutive
positive orbitals of x, and if there is such an orbital, it lies to the right of all non-
trivial orbitals of y in that interval, and the positive orbitals of x, y, z share no
endpoints.
90 J. K. Truss

sameseq(x, y, z, u): the sequences encoded by (x, y) and (z, u) are eventually equal.
In between the element u provides a positive orbital in those intervals between
the positive orbitals of x where z has none. In sameseq, if the positive orbitals of x
and z do have the same left hand endpoints, then we may take the positive orbitals
of v to be given by their intersections, and the condition that yW = u ensures that
the correct ones correspond. The final clause tells us that the left hand endpoints
are indeed equal.
Now it is clear that with gl,h 1 ,g2,h 2 as given, A(nd/L(nI) ~
sameseq(L(nt}.gl, L(nd.hl' L(nI).g2, L(nI).h2)' and since () is an isomorphism,
also A(n 2)/ L(n 2) ~ sameseq(L(n2).g~, L(n2).h~, L(n2).g~, L(n2).h~). Hence {n :
an PI f. a n P2} is finite.

Lemma 5.8 For any a, b there is an isomorphism k from a final segment of n1 to

a final segment of n2 such that {i : ajPa f. aj k } U {i : biPb f. bi k} is finite.

Proof. First we reduce to the case where b is an infinite subsequence of a and

its complement in a is also infinite, where we show that {i : biPa f. bjPb} is finite.
For let c be chosen so that each of a and b are subsequences of c and each has an
infinite complement in c, and let k be an isomorphism from a final segment of n1 to
a final segment of n2 extending Pc. Then by the special case, {i : aiPa f. aiPC}, {i :
biPb f. bipc} are finite, and hence so is {i : aiPa f. aj k} U {i : biPb f. bik}.
To prove the special case we will use a method precisely analogous to that of
Lemma 4.12. Consider the following formula:
stare(x!, X2, ... , x n , y, z) : 3x(x = XIX2 ... Xn /\ inc(y, z)/\
/'\1<r<r'<n disj(x r , x r') /\ link(x, yz) /\ link(yz, x) /\ 3s3t[disj(s, t) /\ x =
=
st i\ link(s, z) /\ link(z, 05) /\ 3s3t[disj(s, t) /\ x st /\ link(s, z) /\ link(z, s)/\
3u(pas(u) /\ disj(u, z) /\ til = y) /\ 3u3v3w(pas(u) /\ pos(v) /\ pos(w)/\
disj(u, y) /\ disj(v, t) /\ disj(w, Xil) /\ X UV = s /\ yW = t/\
/\1 <!:,r<!:,n X~V = S/\ x,.»).
This is intended to store disjoint elements gj whose product is a moiety with
positive orbitals 10 < II < 12 < .... Essentially the intuition is similar to that
behind the construction of the formula inc (though more complicated). Given the
gi we form moieties hi such that gj is a restriction of h j and for each n there are
exactly n + 1 positive orbitals of TI h j between In and I n +!, and they are distributed
among the hi in precisely the same order as the orbitals of TI gi to the left of In+1
are among the gi. Thus TI hj 'remembers' the pattern of the giS cofinally often, and
so this pattern is not destroyed on passing to the quotient by L(nd. To control
recovery of the stored pattern we insert additional markers in the shape of a pair
satisfying inc.
In the formula, Xi stands for L(nd.hi, and y, z for the additional markers; t
stands for L(nl). TI gi and positive elements are used to conjugate TI hi to the right
so as exactly to match s after passing through one of the positive orbitals of y.
Recovering Structures from Quotients of their Automorphism Groups 91

Now to apply this formula we choose a pair (g, h) satisfying inc with ai the
left hand endpoints of the positive orbitals Ii of g, and with Jij for j :s: i the
positive orbitals of h lying between Ii and I i +1 , to define Pa. We use the re-
striction of g to U{ Ii : ai an entry of h} and a corresponding restriction of h
by taking the correct number of JijS for such i as required to define Ph' We
have four restrictions g1, g2, g3, g4 of g h therefore to remember, the restrictions
of g, h just described, and their restrictions to the complement. What matters
is the order these come in. We form elements k and I so that A(n J )/ L(nd F
store(L(nt).gl,"" L(nd.g4, L(nd·k, L(nd·/). As 0 preserves the truth of store
there are gL kl, II such that L(n1).gf =L(n 2 ).g: and so on, and A(n 2 )/ L(n 2 ) sat-
isfies the corresponding store statement. The outcome as in Lemma 4.12 is that
the arrangements of the orbitals of gl, g2, g3, g4 and of g~, g~, g;, g~ are the same. In
other words there is an isomorphism from a final segment of n 1 to a final segment
of n 2 taking gi to g: for each i. Hence {i : biPa I- biPh} is finite, as required.

Theorelll 5.9 Let n 1 , n2 be doubly homogeneous chains having final segments iso-
morphic to R. Then for any isomorphism 0 : A(n 1 )/ L(nd --;. A(n 2 )/ L(n 2 ) there
is an isomorphism from a final segment of n 1 onto a final segment of n 2 which
induces O.

Proof Given the development so far it follows that for any a there is a map Pa
obtained naturally from 0, defined on the set of entries of a and taking a to some
strictly increasing divergent sequence in n2 . The previous lemma showed that any
two such PaS are compatible, and so the remaining task is to show how they can
all be fitted together to form the desired k. In §4 we were able to handle this in
cases where the whole set could be expressed in terms of finitely many (two in fact)
of the approximants. To simulate the same situation here there are various options
available. The most straightforward, as in [5], involves use of finitely many elements
in A(n 1 ) such that there is an orbit of the group they generate which is dense on a
final segment of n 1 .
One of the tricks used in [7] involved consideration of translations g and h of
R such that (g, h) has a dense orbit, for instance xg = x + 1 and xh = x + -../2.
We can envisage copying this behaviour on intervals arbitrarily far to the right in
n 1 , and hence forming a grid which will sufficiently well approximate all points of
a final segment of n 1 .
Choose points {an : n E Z} in n 1 ordered in the natural way and unbounded
above, and elements gl, g2, g3, g4 so that supp(gt) = SUPP(g2) = U{(a2n, a2n+d :
n E Z}, SUPP(g3) = SUPP(g4) = U{(a2n+l, a2n+2) : n E Z}, and for each n there
is an isomorphism from R to (an, a n +1) which carries translation by 1 to gl, g3 (n
even, odd) and translation by.)2 to g2,g4 (n even, odd).
By Lemma 5.8, if a1 = (ao, a2, a4, ... ) and a2 = (aI, a3, a5, ... ), there are bn E n 2
for n E Z such that {n 2: 0 : a2nPa l I- b2n } U {n 2: 0 : a2n+lPa 2 I- b2n +d
is finite. By the techniques already illustrated there are g; E A(n2) satisfying
92 J. K. Truss

the corresponding properties with respect to the bn as gi do with respect to the

an. An important point, which may not be entirely clear, is that there are order-
isomorphisms of R to (b n , bn + l ) which carry translations by 1 and .j2 to g~, g~ and
g~, g~ (depending on n even or odd). The essential reason for this is that .j2 can
be arbitrarily well approximated by rationals, and for any positive integers i and
j, ifj < .j2 ¢:> L(Sld.gi < L(SlI).g~ and similarly for g3 and g4. What this argument
shows is that if an isomorphism is chosen from R to (b 2n , b2n +d which carries
translation by 1 to g~, then g~ arises from translation by Tn where (Tn) is a sequence
with limit .j2. To establish the stronger statement, that {n : Tn ::j:. .j2} is finite
(which is required), we observe that in A(nd/ L(Sld there is an element satisfying
posl which commutes with each of L(Sld.gl and L(SlI).g2, and consequently there
must be a corresponding such element in A(Sl2)/ L(Sl2), and this ensures that (Tn)
is eventually constant.
Finally letting c = (c n : nEw) where Cn E (an, an+d we can now reach
arbitrarily close to any member of (ao, 00) by starting from some C n and applying a
member of L(Sld.(gl,g2,g3,g4)' Letting c~ = CnPc we define k: (ao,oo) -> (bo,oo)
by
ank
(c2ng~gDk
(C2n+lg;g~)k
This defines k on a dense subset of (ao, 00), by the remarks made above it is order-
preserving, and so extends to a map k as required. Moreover, since the whole
procedure is reversible, k is onto.
This tells us how we can choose k. Now we have to show that it has the desired
properties. First, for any e, {i : eiPe ::j:. Ci k} is finite. For this by use of Lemma 5.S,
and without loss of generality, we may restrict to the case where a2n < en < a2n+l
for each n. By Lemma 5.S again, Pe may be chosen so that b2n < e~ < b2n + l for all
n. Now en is determined by {( i, j) : c2ngi g~ < en}, and if this set is the same for
all n (or eventually constant) then there is no difficult.y. To cope with the (usual)
situation where this set is not const.ant, we have once again to 'store' the en as n
increases. This is partly why we ensured t.hat SUPP(gl), SUpp(g2) have unboundedly
many gaps; so that. there is room to fit the st.oring mechanism. We choose moi-
eties hI, h'2 whose support.s are UnE",(C'2n, d 2n ) and UnEw(c n , In), both contained in
UnEw(a2n,a2n+d, whose positive orbitals have no am as an endpoint, and copy the
quadruple (gl, g2, hI, h2) onto the vacant intervals, so that each contains informa-
tion on the behaviour of these elements to the left of this point. This is enough now
to be able to preserve the trut.h or falsity of c2ngig~ < en under e. (For instance
we may say t.hat there is an clement conjugate to h2 by a negative conjugacy with
support disjoint from g3, whose support is contained in that of hi:g~ but is disjoint
from that of h2)' Hence for all i and j, c2ngig~ < en ¢:> cS n (gU i (g~)j < e~, and this
implies that {i : eiPe ::j:. Ci k} is finite.
Recovering Structures from Quotients of their Automorphism Groups 93

It remains to show that k induces the isomorphism e. Since A(Ot}, A(02) are
generated by moieties (at least when we pass to the quotients), it suffices to show
that for any moiety g, (L(01).g)9 = L(02).k- 1gk. Let 9 have positive orbitals
(a2n, a2n+1) for nEw, and choose a moiety g' E A(02) having positive orbitals
(a~n' a~n+1) and such that (L(0I).g)9 = L(02).g'. Now if a2n+1 ::; x ::; a2n+2, xg =
x, and a~n+1 ::; xk ::; a~n+2' so that xkg' = xk = xgk. If there are arbitrarily large x
such that xkg' f:- xgk there must therefore be b such that for each n, (b 2n , b2n+d is
contained in a distinct positive orbital of 9 and b2n kg' f:- b2n gk. Choose a moi-
ety h having support UnEw(b 2n , b2n +d, so that h g is a moiety having support
UnEw(b 2n g, b2n +1g). Let Cn = bng. Then {n : b2nPb f:- b2n k} U {n : C2nPC f:- c2nk}
is finite, and for n outside this set, c~n = C2nPC = C2n k = b2n gk f:- b2n kg' =
b2n Pbg' = b~ng'. But also if (L(0t).h)9 = L(02).h', then for large enough n, h' has
the (b~n' b~n+1) as positive orbitals, and so (h'/ has the (b~ng', b~n+1g') as positive
orbitals. Hence for large enough n, c~n = b~ng'.
This contradiction shows that {x E 0 1 : xkg' f:- xgk} is bounded above, and
hence so is {y E O2 : yg' f:- yk- 1gk}. Therefore (L(0t).g)8 = L(02).k-1 gk, as
required.

Theorem 5.10 Let 0 1 , O2 be doubly homogeneous chains having final segments

isomorphic to Q. Then for any isomorphism e : A(Ol)/ L(Ot} ---> A(02)/ L(02)
there is an isomorphism from a final segment of 0 1 onto a final segment of O 2
which induces e.
Proof. This is proved by a very similar method to Theorem 5.9. Since Q is not
order-complete, there are some additional complications. For instance, there are
four conjugacy classes of positive elements satisfying conj2; two of them consisting
of contiguous ones, where the positive orbitals have either all rational or all irrational
endpoints; and two of them separated, where the positive orbitals are (an, bn ) so
that either all the an are rational (and all the bn irrational), or all the an are
irrational. It is possible to distinguish all four of these classes however by means
of a first order formula. The trick again derives from methods given in [7], and
so provided we choose one and stick to it, (contiguous with rational endpoints for
instance), and make appropriate modifications all through the proof, there is no
insurmountable difficulty.

Theorem 5.11 (i) All automorphisms of A(R)/ L(R) or A(Q)/ L(Q) are znner.
(ii) The outer automorphism groups of A(R)/ B(R) and A(Q)/ B(Q), respec-
tively, have order 2.

The first part is an immediate consequence of Theorems 5.9 and 5.10, and the
deri vat ion of (ii) from these is given in [5].
A part of the material I have described in the section is implicit (in more gen-
erality) in [12]. Namely, Rubin's techniques enable one to interpret various classes
94 J. K. Truss

inside A(R)/ B(R), such as the family of sets of the form U{( a2n" a2n+d : nEw}
modulo sets bounded above, where (an )nEw is a strictly increasing divergent se-
quence, and using his results on locally movable groups the action of A(R)/ B(R)
on the boolean algebra of regular open subsets of R modulo those bounded above
is also interpretable. His techniques do not appear to extend to the part of the
argument where all automorphisms of A(R)/ B(R) are shown to be inner.
One may enquire to what extent the results indicated here for Rand Q hold for
other chains. Moreover, the form of the results is somewhat unsatisfactory, (though
formulated similarly to Theorem 4.13), in that although stating the results in terms
of isomorphisms between different A(n)/ L(n)s makes the results look a little more
natural, the restriction on the allowed values of n is so stringent as actually to
give very little genuinely extra. What would be much more satisfactory would
be a result of the form: any isomorphism () from A(nd/ L(nd to A(n 2 )/ L(n2)
is induced by an isomorphism between final segments of n 1 and D 2 . This is too
ambitious without some further hypotheses on the two chains. With the techniques
so far developed the very least that we would expect to have to assume is that
they are doubly homogeneous and have countable cofinality. In this case, in view
of the 'recognition' results of [7), it seems one can make things at least seem a little
more general than I have done here. The case of uncountable cofinality has not be
touched on, and so a good test case seems to be to look at the long line L. Does
A(L)/ L(L) have any outer automorphisms?

References
[1) J. L. Alperin,.1. Covington and H. D. Macpherson, Automorphlsms of quotients
of symmetric groups, this volume.

[2] R. N. Ball and M. Droste, Normal subgroups of doubly transitive automorphism

groups of chains, Trans. Amer. Math. Soc. 290(1985),647-661.

[3) N. G. de Bruijn, Embedding Theorems for Infinite Groups, Indag. Math.

19(1975),560-569.

[4) U. Feigner and F. Haug, The homomorphic Images of infinite symmetric groups,
Forum Mathematicum, 5(1993), 505-520.

[5) M. Giraudet and J. K. Truss, On distznguishing quotients of ordered permuta-

tion groups, Quarterly Journal of Mathematics (2), 45(1994), 181-209.

[6) A. M. W. Glass, Ordered Permutation Groups, London Math. Soc. Lecture

Notes Series Vo!' 55, Cambridge University Press, 1981.

[7) Y. Gurevich and W. C. Holland, Recognizing the real line, Transactions of the
American Mathematical Society, 265(1981),527-534.
Recovering Structures from Quotients of their Automorphism Groups 95

[8] M. Hall, The Theory of Groups, Macmillan, 1964.

[9] R. Kaye and D. Macpherson (editors), Automorphisms of First Order Struc-

tures, Oxford University Press, 1994.

[10] R. McKenzie, On elementary types of symmetric groups, Algebra Universalis

1(1971), 13-20.

[11] M. Rubin, On the automorphism groups of homogeneous and saturated boolean

algebras, Algebra Universalis 9(1979),54-86.

[12] M. Rubin, The reconstruction of boolean algebras from their automorphism

groups, in Handbook of Boolean Algebras, (edited by J. D. Monk, R. Bonnet),
Elsevier (1989),549-606.

[13] M. Rubin, The reconstruction of topological spaces from their groups of home-
omorphisms, Transactions of the American Mathematical Society 312(1989),
487-538.

[14] M. Rubin, The reconstruction of trees from their automorphism groups, Con-
temporary Mathematics 151(1993), American Mathematical Society.

[15] M. Rubin, Locally moving groups and reconstruction problems, this volume.

[16] W. R. Scott, Group Theory, Prentice Hall Inc, Englewood Cliffs, N.J., 1964.

[17] S. Shelah, First order theory of permutation groups, Israel Journal of Mathe-
matics, 14(1973), 149-162.

[18] S. Shelah, Errata to: First order theory of permutatIOn groups, Israel Journal
of Mathematics, 15(1973), 437-44l.

[19] S. Shelah and J. K. Truss, On distinguishing quotients of symmetric groups, in

preparation.

[20] J. K. Truss, Infinite permutation groups; products of conjugacy classes, Journal

of Algebra, 120(1989),454-493.
 The Automorphism Groups of Generalized McLain
Groups *

Manfred Droste Riidiger Gobel

Institut fur Algebra FB 6, Mathematik und Informatik

Technische Universitat Dresden Universitat GHS Essen
01062 Dresden, Germany 45117 Essen, Germany
e-mail: [email protected] e-mail: [email protected]

1 Introduction
In 1954, McLain [16] applied the well-known connection between nilpotent algebras
(of triangular matrices) and nilpotent groups to establish the existence of partic-
ular locally nilpotent groups, now known as McLain groups. McLain groups are
important for many questions in group theory, and their construction and proper-
ties can be found in most text books of infinite group theory; we refer to [17, pp.
347-349]. The basic ingredients in McLain's construction are the linear ordering Q
of rational numbers and the field Fp of p elements (p a prime), which give rise to
McLain groups G(Fp, Q) which consist of upper triangular "Q x Q-matrices" over
Fp (rows and columns labeled by numbers in Q) with l's on the diagonal and only
finitely many entries from Fp \ {O} elsewhere. These matrices are clearly invertible
and constitute a group G = G(Fp,Q). The field Fp makes G a locally finite p-
group and even if Fp is replaced by a field of characteristic 0, matrix multiplication
ensures that G is locally nilpotent; moreover Q is homogeneous, hence G is char-
acteristically simple - in contrast to finite p-groups. McLain's ingenious, but also
very transparent construction stimulated further research on McLain groups and
their natural extensions. Most notably, J. Roseblade deeply investigated in his Ph.
D. thesis under supervision of P. Hall in 1963 the "internal structure" of G and
determined its automorphism group; see [18]. Moreover, Wilson [19] studied gener-
alized McLain groups G(Fp, S) replacing (Q,~) by another dense linear ordering S.
His extensions of McLain's work are a source for many interesting examples which
illustrate the strength and limitation of recent results in group theory.
·This work is supported by GIF project No. G-0294-081.06/93 of the German-Israeli Founda-
tion for Scientific Research & Development.

97
W. C. Holland (ed.), Ordered Groups and Infinite Permutation Groups, 97-120.
© 1996 Kluwer Academic Publishers.
98 Manfred Droste and Rudiger Gobel

This class of McLain groups can be extended further if the two parameters Rand
S in the construction of McLain groups C(R, S) range over all rings R and arbitrary
partial orderings S; see section 2 for the general definition. Lengthy calculations in
[5] show that Roseblade's characterization of the automorphism group Aut G( R, S)
can be carried over mutatis mutandis to arbitrary rings and orderings if and only if R
has no zero-divisors #- 0 and (5,::;) is locally linear and unbounded; see Theorem 2.2.
Hence we will assume throughout this paper that the not necessarily commutative
ring R has no elements r, s both #- 0 but rs = O. Moreover, we assume for the
rest of this paper that (S,:::;) is unbounded and locally linear, i.e., any interval
[x, y] = {s E 5: x:::; s :::; y} is linear. Domains are well-studied, but also
very particular examples of these rings in question and trees ([4],[6]) are obviously
locally linear. Just for convenience we will also suppose that the ordering S is
connected, i.e., any two points of 5 can be connected with a finite path with any
two consecutive points comparable; see section 2 for details. For such pairs of rings
R, R' and orderings S, S' we have shown in [5] that the McLain groups G(R, S)
and G( R', S') are isomorphic if and only if the rings R, R' and the orderings 5, S'
are either isomorphic or anti-isomorphic respectively. This essentially says that
the ring structure and the structure of the ordering can be reconstructed from the
group structure of G(R, S). The answer to this isomorphism problem provides the
foundation to the following more general question on automorphism groups Aut G,
which is the topic of this paper.
( ) { Let C, C' be two McLain groups and suppose that
* Aut C == Aut G'. Can we deduce that C == G'?
This is a well-known kind of question with many ancestors. It was answered for the
classical McLain groups C(Fp, ([Jl) by Roseblade [18]; we will also take the oppor-
tunity to fix the flaw in the proof of the crucial Theorem 3 in [18, p. 275] spotted
by Falkenberg [11]. (*) also has been answered for other structures than McLain
groups in various contexts. The most fundamental relatives of (*) are Dieudonne's
results [2] and [3], showing that isomorphisms CL(F,n) == CL(F',n') of general
linear groups induce isomorphisms of the underlying vector spaces, hence the fields
F, F' are isomorphic and the dimensions n, n' (which may be infinite cardinals) are
equal. A similar result holds for abelian p-groups C, C l The case p 2: 5 is due to H.
Leptin [14] and may be viewed as a type of Ulm 's theorem, and the more recent case
p = 3 is in Liebert [15]. The case p = 2 still remains open. If C, C' are torsion-free
modules over the p-adic integers, then the positive answer to (*) is the main result
of the forth coming paper [1] by Corner and Goldsmith. Many more relatives of
(*) and positive answers may be found in the literature, e.g. replacing "Aut" by
"End", the endomorphism ring (monoid) of the underlying structure C. We want
to add the positive answer for McLain groups. Our main result (Main Theorem
5.8) will be an isomorphism G == G' deduced from Aut C == Aut C' for two McLain
groups G, C'. We will have to require that the rings R used for constructing McLain
groups have the additional property that their ring-automorphisms are determined
Automorphism Groups of Generalized McLain Groups 99

by their action on the group of units U(R) of the ring. This is not very surprising,
because it is U(R) which plays a central role in decoding R, S from Aut G(R, S).
Examples of such rings are all (skew) fields, but also many commutative rings in-
cluding Z. We also note that the isomorphism between G and G' is not induced
from the isomorphism between Aut G and Aut G'. Finally we want to discuss how
to find an isomorphism between G and G'. The strategy is based on a sequence of
subgroups of Aut G, which we explain first:

G = Inn G ~ 1 + pT ~ j3(l +T*) ~ j3(Aut G) ~ o-(G) ~ 1 +T* = Linn G.

As G has a trivial centre, we may identify inner automorphisms (= conjugation by

group elements) with elements in G, hence G = Inn G ~ Aut G. While elements in
=
G may be written in the form g 1 + t with t =L aOi/JeOl/J, aOi/J E R, a, j3 E Sand
OI</J
= 0 for almost all a, j3 E S and multiplication of S x S-matrices eOi/J (having
aOi/J
1 at (a,j3) and 0 elsewhere) is given by eOi/J . e'Y~ = 8/J'Y . e 6 (8/J'Y the Kronecker
0l

symbol), some of the automorphisms of G can be expressed similarly: If T denotes

the ring of all zero-triangular S x S-matrices over R which are row- and column-
finite, then 1 + T forms a semigroup with respect to multiplication and we consider
the maximal group 1 + T* contained in 1 + T. It is easy to show that t E T* if
and only if there exists t* E T with t + t* + tt* = t + t* + tOt = O. The arguments
given by Roseblade [18, pp. 269,270] show that 1 + T* consists of locally inner
automorphisms of G, hence 1 + T* ~ Linn G, where Linn G denotes the group of
locally inner automorphisms of G. Like in the case of classical McLain groups it
can be shown [5] that 1 + T* = Linn G and the matrices in question can be viewed
group theoretical, cf. Corollary 2.3. We deduce that the Baer radical j3( 1 + T*)
must be contained in the Baer radical j3(Aut G), cf. Corollary 2.4. Recall that
Linn G is normal in Aut G and the Baer radical j3H of a group H is the subgroup
of H generated by all subnormal nilpotent subgroups (hence by all subnormal cyclic
subgroups) of H. Following [18] we also see that 1 + pT is contained in the Baer
radical, cf. Lemma 2.5. Here pT denotes the Jacobson radical of T comprising
all matrices contained in some two-sided nilpotent ideal of the ring T. Obviously
G ~ 1 + pT and trivially j3(Aut G) is contained in the subgroup o-(G) of Aut G
comprising all Baer automorphisms of G. Just note that 0:" E Aut G is a Baer
automorphism of G if and only if (0:") is subnormal in the subgroup (G, 0:") of Aut G.
The above sequence of subgroups of Aut G is established after we have shown that
0-( G) ~ Linn G, which is one of the main tasks of this paper; see Theorem 4.1. For
special rings and orderings R, S some of the inclusions of the above sequence are
equalities, which however is not needed for our purpose. If <p is now an isomorphism
between Aut G and Aut G' for two McLain groups G = G(R, S) and G' = G(R', Sf),
then j3(Aut G) is mapped onto j3(Aut G') because the Baer radical is a characteristic
subgroup of Aut G. In order to deduce isomorphisms or anti-isomorphisms between
the rings R, R' and orderings S, S' respectively we now can use our results from
section 3 concerning G-normal closures of subgroups of Linn G. We conclude with
100 Manfred Droste and Riidiger Gobel

(3.3), (3.10), and (3.13) that the McLain groups G(R, S) and G(R', S') must be
isomorphic as well; see Main Theorem 5.8.

2 Automorphisms of the McLain groups

In this section we recall the definition of generalized McLain groups and the de-
scription of their automorphism groups derived in [5].
Let (S,::;) be a partially ordered set (poset) and R an arbitrary ring with 1 f O.
Let E = E(R,S) be the free R-module with basis {e a !3 ; a,f3 E S, a < f3}.Then
E becomes an associative R-algebra by defining a multiplication of basis elements
via

ea6 if f3 = 'Y
eo:!3 . e"(6 ={ 0 otherwise.

Clearly, all elements of E are nilpotent. Hence the set G = G(R, S) of formal
sums 1 + u (u E E) forms a multiplicative group (see, for example [16]). The
= =
general element of G is expressible uniquely in the form 9 1 + a 1 + I: ao:!3e a!3
a<!3
with ao:/J E Rand ao:/J = 0 for almost all a, f3 E S. Subsequently, for short, we
will simply write 9 = 1 + I: ao:/Jeo:!3' We write [g] =
{(a, (3); aa/J f O}, the
support of g, [gh = {a E S ; 3 f3 E S (a,f3) E [g]}, the I-support of g, and
[gh = {f3 E S : 3 a E S (a, (3) E [g]}, the 2-support of g.
We also write [a] = [g] and [a]; = [gl; (i = 1,2), where 9 = 1 + a as above. For
any subset A ~ S, we let max A denote the set of all maximal elements of A. We
say that the poset (S,::;) is dense, if for any a, f3 E S with a < f3 there is 'Y E S
such that a < 'Y < f3.
The following lemma is a straightforward generalization of one due to McLain
[16, Lemma 1]. The representation of an arbitrary group element as a product as
described below will be crucial throughout.

Lemma 2.1 G is generated by the set of all elements 1 + aeo:/J, (a E R, a, f3 E S,

with a < (3).

Next we turn to the description of the automorphisms of G(R, S). As shown in

[5], this requires natural assumptions on Rand S, which we introduce first.
A poset (S,::;) is called unbounded, if for each U' E S there are a, f3 E S with
a < U' < f3, and locally linear, if for any a, f3 E S with a < f3 the set {U' E S: a <
U' < f3} is linearly ordered. Two elements a, f3 E S are called connected if there
are elements aa, . ", an E S such that aa = a, an = f3, and for each 0 :::; i < n,
either ai ::; ai+1 or ai+l ::; a;. Such a sequence aa," " an is also called a path in
S; if here aa = an, we say this path is closed. A component of (S, ::;) is a maximal
connected subset of S. Clearly, S is the disjoint union of its components. For each
component C of S, let G c = {g E G: [g] ~ C x C}. Clearly, G is the direct
Automorphism Groups of Generalized McLain Groups 101

sum of the McLain groups Gc ~ G(R, C) (C a component of S). Therefore it is

natural to assume that the poset S is connected. Furthermore, the description of
the automorphisms of G(R, S) in [5] is based on an analysis of the maximal abelian
normal subgroups of G. As shown in [5], they can be described order-theoretically
if and only if R has no zero-divisors "# 0 and (S,~) is unbounded and locally linear.
Therefore subsequently we will put the following restriction on Rand (S, ~):

R has no zero divisors "# 0, and

(*) { (S,~) is an unbounded locally linear connected poset.
We fix Qo E S throughout.

Note that this assumption is satisfied trivially for the classical McLain groups
where R is the field with p elements (p prime) and S = Q, the rationals, but also
for many other rings and posets.
First we describe automorphisms of G induced by the ring and the order. Let
ip be an order-automorphism of Sand 1/J a ring-automorphism of R. For

Clearly, <I> is an automorphism of G, called the automorphism induced by (ip,1/J).

Next assume that ip is an anti-automorphism of (S,~) and 1/J an anti-automorph-
ism of the ring R. If

let

If
h = 1 + Lb-y6e-y6 E G,

we have

so
(g . h )<I>' = (h<I>') . (g<I>').
Since inversion is a group-anti-automorphism, it follows naturally that <I> defined
by the composition g<I> = (g-1 )<I>' (g E G) is a group-automorphism. We call <I> also
the group-automorphism induced by (ip,1/J).
Now if ip E Aut(S, ~), let ip: G -+ G be the automorphism induced by (ip, id R ).
Similarly, if'IjJ E Aut R, let ;f : G -+ G be the automorphism induced by (ids, 'IjJ). If
Sand R are anti-isomorphic to themselves, choose anti-automorphisms ip* of (S, ~)
and 'IjJ* of R and let T be the automorphism of G induced by ip* and 'IjJ*. Otherwise,
102 Manfred Droste and Rudiger Gobel

we let 7 = 1. Next let (C a,6)a<,6 be a family of units of R satisfying the following

identities:
Ca,6 . c,6",! = ca"'! for all 0:' < (3 < , I and
(**)
{
IIo caia ,+l
n-l
E «(R) for each closed path 0:'0,"', O:'n in S;
here «(R) denotes the centre of R.
We put 'l/;ao = id R , the identity on R. If (3 E S, choose a path 0:'0, .. " O:'n from
n-l
0:'0 to O:'n = (3, put c = I1 caiai+l' and define '1/;,6 by '1/;,6 = c- 1 . r· c. Clearly, '1/;,6 is a
;=0
well-defined automorphism of R by (**), and 'I/;~"'P = '1/;,6 for all 0:' < (3 in S, where
for r E R, r('I/;~"''') = c~J . (r'l/;a) . ca,6·
Now define b : G -+ G by (1 + 'L, aa,6e a,6)b = 1 + 'L,(aa,6'1/;a) . ca,6e a,6. It is easy
to check that b is an automorphism of G, called the dilatation induced by the family
(ca,6)a<,6 and 0:'0·
We note that anyone-parameter family (da)aES of units of R gives rise to a
family (c a,6 )a<,6 satisfying property (**) by putting ca,6 = d;;; 1 . d,6. Conversely, for
a large class of posets (S, ::;) (called loop-free in [5]; examples include all trees) any
family (c a,6)a<,6 arises in this way from a one-parameter family (da)aES'
We let S comprise all automorphisms (j5 of G induced by (cp, idR ) for some cp E
Aut(S, ::;); similarly R = {~ : :;j; induced by (ids, '1/;), 'I/; E Aut R} and D = {b : b
induced by some family (C a,6)a<,6 as above }. Clearly, Sand R are subgroups of
Aut G, isomorphic to Aut(S,::;) and Aut R, respectively. As shown in [5], D is also
a subgroup of Aut G , called the dilatation group. Finally, an automorphism of Gis
called locally inner, if it acts by conjugation on all finitely generated subgroups of
G. Let Linn G comprise all locally inner automorphisms of G; obviously Linn G
is a normal subgroup of Aut G.
Now we can state the central result of [5].

Theorem 2.2 [5, Theorem 6.2] Let R be a ring with no zero-divisors I: 0, and let
(S,::;) be an unbounded locally linear connected poset. Let G G(R, S). Then we =
have the following semidirect product decomposition of Aut G:
Aut G = (((Linn G ><I D) ><IR) ><IS). (7), where 72 E R x S
and
(Linn G ><I D) ><I R is normal in Aut G.

For any 0:' < (3 in S, let Xa,6 comprise all elements x E E such that any pair
(" b) E [x] satisfies, ::; 0:' < (3 ::; b and either, < 0:' or (3 < b. It is easy to see that
if 'I/; E Linn G and a E R, then (1 + ae a(3)'1/; = 1 + ae a,6 + x for some x E X af3.
We will also need another description of locally inner automorphisms. For this,
let M be the free R-module with basic elements ma (0:' E S). For each A E S, let
M>. be the submodule of M generated by all the elements miJ with A ::; Il, and
let M>.+ be the submodule generated by {miJ: A < Il}. Let E+ be the ring of
Automorphism Groups of Generalized McLain Groups 103

those endomorphisms of AI which map M).. into M)..+ for each ,X E S. If 1} E E+

and'x E S, then m)..1} is a finite linear combination of certain of the mJ.' (,X < /1).
Therefore E+ can be thought of as a ring of row-finite zero-triangular matrices, and
defining m)..e)..J.' = mJ.' and mre)..J.' = 0 for T # ,x, E appears as the subring of E+ of
all zero-triangular matrices with almost all coefficients zero. Let T be the subring
of E+ of all matrices which are also column-finite. Then 1 + T forms a semigroup
with respect to multiplication. As before, the elements 1 + t can be expressed as
1 + L: ta{3ecr{3 which, however, now may have infinitely many entries tcr{3 # O. Let
cr<{3
1 + T" be the maximal group contained in 1 + T. Then 1 + T" contains all elements
of 1 + T with a two-sided inverse in 1 + T, and t E T* if and only if there exists
t* E Twith t+t*+tt* = t+t*+t*t = O. Now ift E T*, we obtain an automorphism
1/Jt of G induced by transforming with 1 + t. That is, if 1 + t* is the inverse of 1 + t
in I+T*, we have (l+l'e)..J.')1/Jt = (l+t*)(1+re)..J.')(I+t) for all r E R and'x < /1 in
S. We let Ll comprise all such automorphisms 1/Jt (t E T*) of G. Subsequently, we
identify L1 = 1 + T*. The argument given by Roseblade [18, pp. 269, 270] shows
that Ll ~ Linn G.
Actually we have:

Corollary 2.3 [5, proof of Theorem 6.2]. Let R be a ring with no zero-divisors # 0,
and let (S,:S;) be an unbounded locally linear connected poset. Then L1 = Linn G.

As a consequence we obtain:

Corollary 2.4 Let R be a ring with no zero-divisors # 0, and let (S,:S;) be an

unbounded locally linear connected poset. Then f3L 1 ~ f3(Aut G(R, S».
Proof. By Corollary 2.3, we have Ll = Linn G, a normal subgroup of Aut G.
Hence each subnormal subgroup of Ll is subnormal in Aut G, and the result follows
by the definition of the Baer radical.
We let pT denote the Jacobson radical of T comprising all elements contained
in some two-sided nilpotent ideal of T. Clearly E ~ pT and so G = 1 + E ~ 1 + pT.
If t E pT, we have t n = 0 for some n, hence (1 + t)-1 = 1 - t + t 2 - ... ± t n - 1 and
so t E T*. This shows that pT ~ T* and 1 + pT ~ 1 + T* = L 1 . Furthermore, we
have:

Lemma 2.5 Let R have no zero-divisors # 0 Then 1 + pT is contained in the Baer

radical of 1 + T*. If p = chal'(R) # 0, then 1 + pT is a locally finite p-group. If
char(R) = 0, then 1 + pT is torsionfree.

Proof. By inspection of [18, p. 269, proof of Lemma 1]. The last two statements
are clear.
We will identify elements of G(R, S) with the inner automorphisms induced by
conjugation. Hence G ~ L 1 . As for elements of G, for t = L: tcr{3ecr{3 E T* we
0I<{3
104 Manfred Droste and Rudiger Gobel

write [1/)t] = [t] = {( 0', /3): taf3 i- O}, the support of 1/'t, [1/'th = [th =
{O' E S::3 (3 E S (0',(3) E [tD, ["pth = [th = {(3 E S::3 0' E S (0',(3) E [tl).

3 Abelian normal subgroups of Dinn( G)

In this section, we show that the maximal abelian normal subgroups of Linn( G)
are of a specific form defined order-theoretically. First we give a characterization
of abelian normal closures of cyclic subgroups of Linn( 0). For this, the following
condition first used in Roseblade [18, p .271] will be crucial.
We will say that an arbitrary a E T satisfies the condition (p) if a 2 aba =
0 =
for any bEE.

Theorem 3.1 Let (5,::;) be unbounded. Let 9 = 1 + a be an arbitrary element of

Linn 0 and N = (gG) be the G-normal closure of 9 in Linn G. The following are
equivalent:

1. N zs abelian.

2. a satzsfies the condition (p) in T.

Proof. Clearly, N is abelian if and only if all generators of N commute, and this
is the case precisely if for all h E 0, gh and 9 commute. With 9 = 1 + a and a E T,
this is equivalent to

h -J . a . h . a =a .h -1 . a . h for all h EO.

Clearly condition (p) implies that aha = 0 for all h = I + b E 0, and hence
(*) holds. Conversely assume (*) holds. Suppose a 2 i- O. Let h = 1 + e a (3 such
that 0' E [a 2 h and (3 tJ. [all. Then h- 1 =
I - e a f3, so h- 1 aha =
a 2 and ah- 1 ah =
a +a ·e a f3 i- a , contradicting (*). Hence a = O. Now suppose there is bET with
2 2 2 2

aba i- O. Then there are pairs (0', (3), (,,8) E [a] and ((3, ,) E [b] and coefficients
x E R of e(1(3 in a and y of ef3-y in b, and z of e-yb in a such that x . y . z -#- O. Put
h = 1 + ye(3-y + ebT! for some TJ tJ. [ah U [ah. Then 11- 1 = 1 - yef3-y - eDT!. Using
a2 = =
0, we obtain h-laha aha = a· ye(3-y . a; but ah- 1 ah = -a· yef3-y . a· h which
has the non-zero coefficient -x . y . z of eat/, again contradicting (*). The result
follows.
Next we derive a commutator formula for elements in 1 + T*.

Lemma 3.2 Letg = l+a= I+Laaf3eaf3 E I+T* andg- 1 = l+a*. Let, <~ <
« 8 and u = l+ze(b, v = l+ye-y~. Then h:= [g,u] = 1-a*ze(b -(l+a*)ze(ba
and [h, v] = 1 - ya~(ze-yb + ye-yf.a* ze(Ij(l. If 9 belongs to an abelian subgroup N
of 1 + T* such that G is contained in the normalizer of N in 1 + T*, we have
[h, v] = 1 - ya~( ze-yb·
Automorphism Groups of Generalized McLain Groups 105

Proof. Using a + a* + a*a = 0, we obtain the formula h = g-lu- 1 gu = 1 + b

with b = -a* ze(b - (1 + a*)ze(6a. Similarly, [h, v] = 1 - (1 + b*)ye'Y€b - b*ye'Y€'
Under the additional assumption, by condition (p) of Theorem 3.1 we have a 2 =
b2 = 0, hence a = -a* and b = -b*, and a*ca = b*cb = 0 for all c E E. Thus
[h, v] = 1 - ye'Y(b + bye'Y€ = 1 - ye'Y(b = 1 - ya€( ze'Y6, as claimed.
We introduce some further notation. Let X ~ S. We define IX = {s E S :
s ~x for some x E X} and TX dually. We call X downward (upward) closed if
X = IX (X = TX), respectively. For singletons XES, let Ix = !{x} and define
Tx dually. We say that X is upper directed if for any x, y E X there is z E X
with x, y ~ z; lower directed is defined dually. X is an (order-) ideal of S, if it is
non-empty, downward closed and upper directed. Dually, X is an (order- )filter of
S, if it is non-empty, upward closed and lower directed. Let A, B ~ S. We write
A < B if a < b for all a E A, bE B; and abbreviate A < {x} to A < x. We call
(A, B) a couple, if the following three conditions hold:

1. A, B are non-empty and disjoint, and A is downward and B upward closed;

2. for all a E A there is bE B with a ~ b;

3. for all bE B there is a E A with a ~ b.

We let C comprise all couples (A, B) of (S, ~).

If A is an ideal, B a filter on S and A < B, we call (A, B) an ideal couple. Let

C* consist of all ideal couples of (S, ~). We introduce a partial order on C by letting
(A, B) ~ (A', B') if A ~ A' and B ~ B'. We write J <I R to denote that J is a
two-sided ideal of R. We also let <I denote normal subgroups; the intended meaning
will be clear from the context. If (A, B) is a couple and 0 f:. J <I R, we call (A, B, J)
a triple. If, here, (A, B) is an ideal couple, we call (A, B, J) an ideal triple. We
put (A,B,J) ~ (A',B',J') if(A,B) ~ (A',B') and J ~ J'. Each triple (A,B,J)
induces a subgroup Ni,B which consists of all elements of 1 + T* of the form

9 =1+L x"'l'e",1' with x"'l' E J, a E A, f3 E B.

Hence [gh ~ A, [gh ~ B. We will write NA,B = Nf B and call these groups
order-theoretic normal subgroups of 1 + T*. '

Lemma 3.3 Let (S,~) be unbounded.

(a) For any triple (A, B, J), Ni B is an abelian normal subgroup of L1 contained
in 1 + pT. '

(b) For triples (A, B, J), (A', B', J') we have (A, B, J) ~ (A', B', J') if and only
if Ni,B ~ Ni:,BI.
106 Manfred Droste and Rudiger Gobel

Proof· (a) Clearly Nj",B is abelian. Let l+a E Ni,B and (l+t) ELI with t E T*
as described at the end of section 2. Then (1 + a)U+t) = (1 + t·)(1 + a)(l + t) =
1 + t* + a + t + t·a + at + t*t + t*at =
1 + a + t·a + at + t*at E Nj" B ' since
[t·a)' [at), [t·at] ~ A x B. So Nj" B <J L 1 . Since a 2 = 0, we have a E ;T and so
Nj",B ~ 1 + pT. '
(b) Let NAJ B C NAI
,-
J'
B', =
, .
N , and a E A, j3 E B, x E J Wlth a < j3 and
n
X 1 O. Then 9 = 1 + xe a (3 E N', so it can be expressed as 9 = 1+ L aje a .(3,.
jEI
The uniqueness of this representation shows I = {I}, say, and thus x = aj E J',
a = =
aj E A', j3 j3; E B' as desired.
Next we show that certain abelian subgroups of 1 + T* are covered by order-
theoretic normal subgroups.

Theorem 3.4 Let (5,~) be an unbounded locally linear connected poset and R a
ring with no zero-divisors =F O. Let N be an abe/ian subgroup of 1 + T* such that
the normalizer of N contains G. Then N is contained in an order-theoretic normal
subgroup of 1 + T· .

Proof. Given N as above, define A = =

UgEN[gh and B U9EN[gh- We claim
that (A, B) is a couple; it will follow that N C N: B. We first show that A = lAo
9
Let a' E 5 and a E A with a' < a. There are E Nand j3, j3' E 5 with (a, (3) E [g),
and j3 < j3'. Then a' E [g'h for some g' E N by Lemma 3.2. Similarly B = lB.
Next we observe that there are no elements g, hEN and a, j3", 8 E 5 with
(a, (3) E [g), (,,8) E [h] and a < j3 < , < 8 in 5. For suppose, otherwise. By
assumption on (5, ~), there is a' E 5 with a' < a and (a',8) ¢ [h]. By Lemma
3.2, for some 0 1 a E R, the element g' = 1 + ae a l , belongs to N. Since R has no
zero-divisors 1 0, we have (a',8) E [g'h]; but clearly (a', 8) ¢ [hg'), so g' and h do
not commute, a contradiction.
Now we show that An B = 0. Otherwise, there are g, hEN and some ~ E
[gh n [hh- Write 9 = 1 + a, h = 1 + b with a,b E T*. Choose a,j3 E 5 with
(a,O E [g), (~, (3) E [h]. By the observation of the preceding paragraph, ~ is
a maximal element of [hh and a minimal element of [gh- Now if , E 5 with
(a,,) E [g] and (" (3) E [h), then , ~ ~ or ~ < , by local linearity of 5, and so
, = ~ by the maximality, respectively, minimality property of ~. Since R has no
proper zero-divisors, we have (a, (3) E lab]. However, gh = hg implies ab = ba.
Thus there is 8 E 5 with (a, 8) E [b] and (8, (3) E [a]. Again, since 5 is locally linear,
we have a < 8 ~ ~ < f3 or a < ~ < 8 < f3. Then () = ~ by the observation noted
above. Thus (a,O, (~,f3) E [a]. By the preceding argument (with a = b), we get
(a, (3) E [a 2 ]. So a2 1 0, contradicting Theorem 3.1. Hence (A, B) is a couple.
Now we can characterize group-theoretically the maximal order-theoretic normal
subgroups of subgroups U of 1 + T* containing 1 + pT. Their forms turn out to be
independent of the choice of U.
Automorphism Groups of Generalized McLain Groups 107

Corollary 3.5 Let U be an arbitrary subgroup of 1 + T* with 1 + pT ~ U, and let

N ~ U. Then the following are equivalent:
1. N is maximal with respect to being abelian and containing G in its normalizer.
2. N is a maximal abelian normal subgroup of U.
3. N = NA,B for some maximal couple (A, B).
Proof. Subsequently we use that by Lemma 3.3(a) all order-theoretic normal
subgroups NA,B are contained in 1 + pT ~ U. We first show that either of (1) or
(2) implies (3). By Theorem 3.4, N is contained in some order-theoretic normal
subgroups N A,B of 1 + T*. Then N = N A,B by maximality of N. By Lemma 3.3,
it follows that (A, B) is a maximal couple. Next we show that (3) implies (1) and
(2). As noted before, NA,B is an abelian normal subgroup of 1 + T* and hence of
U. If N ~ M ~ 1 + T*, the normalizer of M contains G, and M is abelian, then
M ~ NA',B' for some couple (A',B') by Theorem 3.4. Hence (A,B) ~ (A',B') by
Lemma 3.3. Thus (A, B) = (A', B') and N = M.
In order to strengthen the link between the poset of order-theoretic normal
subgroups of G(R, 5) and the underlying poset (5, S;), we need to investigate de-
compositions of couples into ideal couples. We say that two couples (I, F), (I', F')
are disjoint, ifthere is no (a, {3) E (I n 1') x (F n F') with a < {3j equivalently, there
is no couple (I*, F*) contained in both (I, F) and (I', F'). Then, in particular,
In F' = 0 = I' n F. Also, two ideal couples (I, F), (I', F') are disjoint iff either
I n l'= =
0 or F n F' 0.
Now let A = {(Ij, Fj): j E J} ~ C be a family of pairwise disjoint couples,
and let A = UEJlj, B = UjEJFj. Then (A,B) E C. We will write (A,B) =
EBjEJ(Ij,Fj) if (Ij,Fj) ~ (A,B) for each j E J and for all a E A, {3 E B with
a < {3 there is j E J with a E I j , {3 E Fj. Furthermore, if 0 =I X <l R is a non-trivial
i
ideal, we let N comprise all elements 9 = 1 + T* of the form 9 = 1 + 2: x ai3eai3
such that for each (a, {3) E [g) we have xa{3 E X and (a, {3) E I j x Fj for some j E J.
Then Ni ~ N1,B' and we have equality iff (A,B) = EBjEJ(Ij,Fj ). In any case,
arguments similar to Lemma 3.3(a) show that Ni is an abelian normal subgroup
of L1 contained in 1 + pT. We put NA = N!l.
Next we show that order-theoretic normal subgroups have a well-behaved de-
composition property similar to the famous Azumaya-Krull-Remak-Schmidt result
on modules having summands with local endomorphism rings (cf. [12)).
Theorem 3.6 Let (A, B) E C and A = {(Ij , Fj): j E J} ~ C such that (A, B) =
tf}jEJ(Ij,Fj ). Let 0 =I X <l R be a non-trivial ideal.

(a) If A = Al U A2 (disjoint union), then N1,B = Nil EB Ni2 •

(b) Let U ~ 1 + T* be an arbitrary subgroup with 1 + pT ~ U. Let A ~ C* and
assume Nt B = N1 EBN2 with normal subgroups N l , N2 <l U. Then there exists
a decompos'ition A = =
Al U A2 such that Ni Ni, for i 1,2. =
108 Manfred Droste and Rudiger Gobel

Proof· (a) If 9 = 1 + I: xa{Jea{J E Ni.B and (0',13) E [g], then (a,f3) E Ij x Fj

for some uniquely determined j E J. For i = 1,2, let gi = 1 + I: Xa{J ea{J , meaning
A.
that each (a, 13) E [g;] belongs to some Ij x Fj with (Ij, Fj) EA. Then 9 = gl . g2 E
Ni, . Ni2 •• Clearly Ni, n Ni2 = {I}.
(b) For i = 1,2 let A comprise all couples (1, F) E A for which there are
a E I, 13 E F and 9 E Ni with (a, 13) E [g]. Fix some 01:- x E X. We first show that
A = Al uA 2 • Let (1, F) E A and choose a E I and 13 E F. Then 1 + xea{J = gl . g2
with gi E N i , in particular [gil ~ A x B (i = 1,2). Hence (a,f3) E [gil for some
i E {I, 2}, so (1, F) E Ai·
Next we show that Al and A2 are disjoint. Suppose there was (1, F) E Al nA 2 .
For i = 1,2, choose gi E Ni and (ai,f3;) E [gil with aj E I, f3i E F. Let rj E R
be the coefficient of ea .f3. in the representation of gi = 1 + I: l'a{Jea{J. There is
a E I and 13 E F with {al,a2} :S a < f3:S {f31,f3d· Then 1 + xea{J E NiB'
so 1 + xea{J = hI . h2 for some hi E Ni (i = 1,2). As above, (a,f3) E [hd U [h 2],
and we may assume that (a, 13) E [hI]' Let q E R be the coefficient of ea{J in
the representation of hI = 1 + I: q(lTe'lT' Now choose /,8 E S with / < a2 :S
a < 13 :S 132 < 8. Then q . 1'2 1:- 0 by assumption (*) and by Lemma 3.2 we have
1 1:- 1 + q . r2e-yc E Nl n N 2 , a contradiction.
By (a) and A = Al U A2 we have N i.B = Ni, EB N i 2 • Clearly, by definition of
A, each element of Nj belongs to Ni, (i = 1,2). Hence Nj = Ni, (i = 1,2).
This result motivates further investigation of C· and decomposition of couples
into disjoint sums of ideal couples. The following lemma was shown in [5, Lemma
3.3].

Lemma 3.7 Let (1,F),(1',F') E CO with InI' 1:- 0 1:- FnF'. Then I ~ I' or
I' ~ I. Also, F ~ F' or F' ~ F. Moreover, if I\I' 1:- 0, there is a E I with I' < a,
and if In F' = 0, then 1< F'.

The following was shown in [5, Proposition 3.4, Lemma 3.6].

Proposition 3.8
(a) Let (A, B) E C and J ~ C'. The following are equivalent:

(l) (A, B) = EB(I,F)EJ(1, F).

(2) J comprises all ideal couples (1, F) which are maximal with respect to (1, F) ~
(A, B).
(b) Let (A, B) E C be maximal in C, and let (1, F) E C· be a maximal ideal couple
with respect to (1, F) ~ (A, B). Then (I, F) is maximal in C*.

For the rest of this section,

let U ~ 1 + T· be an arbitrary subgroup with 1 + pT ~ U.
Automorphism Groups of Generalized McLain Groups 109

Let N <I U. If N = Nl EB N2 with normal subgroups N l , N2 <I U, we call N;

direct summands of N, denoted by Ni ~ N. If N has no non-trivial proper direct
summand, then N will be called indecomposable in U.

Corollary 3.9 Let (A, B) E C and 0 =F X <I R. Then NiB is indecomposable in

U if and only if (A, B) E C*. '

Proof Assume NiB is indecomposable. By Proposition 3.8(a), (A, B) =

F) for som~ subset :7 ~ C*. Then by Theorem 3.6(a) we get 1:71 = 1,
Ei3(I,F)E.J(I,
so (A, B) E C*. The converse is immediate by Theorem 3.6(b) with A = {(A, Bn.
Recall that Corollary 3.5 gave a group-theoretic characterization of the order-
theoretic normal subgroups NA,B where (A, B) is maximal couple in C. Now we
characterize the normal subgroups NI,F for which (I, F) is a maximal ideal couple in
C*. This provides a strong link between the order-structure of (S, :::;), represented
by (C*, :::;), and the poset of abelian normal subgroups of U. We will apply this
result to the Baer radical of Aut G, which will be shown later (see Corollary 4.2)
to be contained in 1 + T* .

Theorem 3.10 Let U ~ 1 + T* be an arbitrary subgroup with 1 + pT ~ U, and let

N <l U. The following are equivalent:

(1) N = NI,F for some maximal ideal couple (I, F) in C*.

(2) N is abelian, indecomposable and a direct summand of some maximal abelian
normal subgroup M <l U.

(3) N is abelian, indecomposable and a direct summand of any abelian normal

subgroup of U containing it.

Proof (1) - l - (2). By Lemma 3.3 N is abelian, and by Corollary 3.9 it is inde-
composable. By Zorn's lemma, there is a maximal couple (A, B) in C containing
(1, F). Then M = NA,B is a maximal abelian normal subgroup of U by Corol-
lary 3.5. Trivially, (I, F) is maximal in C* with respect to (I, F) ~ (A, B). By
Proposition 3.8(a), we have (A, B) = Ei3(I',F')E.J(I', F') for some :7 ~ C* with
(I, F) E:7. Let:71 {(I, = Fn
and :72 = :7\:71. Then Theorem 3.6 implies
N = N.J1 ~ N.J1 EB N.J2 = M.
(2) - l - (1). By assumption, we have M = NEB J{ for some J{ <I U. By Corollary
3.5 and Proposition 3.8, we obtain M = NA,B = Ei3(I,F)E.J NI,F for some maximal
couple (A, B) in C and some subset :7 ~ C* containing all ideal couples (I, F) which
are maximal in (A, B). The exchange property, Theorem 3.6 implies that N = N.J'
for some:7' ~ :7. Then 1:7'1 = 1, since N is indecomposable. So, N = N(I,F) where
:7' = {(I, F)}, and (I, F) is maximal in C* by Proposition 3.8.
(2) - l - (3). Clear by using the modular law.
(3) --> (2). Apply Zorn's lemma.
110 Manfred Droste and Riidiger Gobel

Next we characterize "principal" couples of the form (la, j,B), (a,,B E S, ci <,B)
by intersection properties of maximal ideal couples.

Lemma 3.11 Let (I, F) E C* be an ideal couple.

(a) There exist maximal ideal couples (Ij, Fj) E C' (j = 1,2) such that I = h n 12
and F = Fl n F 2 ; they coincide iff (I, F) is a maximal ideal couple.
(b) The following are equivalent:

(1) (I, F) = (la, j,B) for some a,,B E S with a < ,B.
(2) There are at least two distinct ideal couples (Ij, F j ) E C* (j =
1,2)
maximal with respect to being properly contained in (I, F) such that
h n 12 =I- 0 and Fl n F2 =I- 0.
Proof. (a) Put h = I and, by Zorn's lemma, let Fl be a filter in S maximal
with F ~ Fl and I < Fl. Assume that. (I', FI) is an ideal couple with h c I'.
By Lemma 3.7, there is a E I' with h < a, so (h, ja) is an ideal couple with
Fl C j a, contradicting the maximality of Fl. Hence (h, F l ) is a maximal ideal
couple. Dually, let 12 be an ideal in S maximal with I ~ 12 and h < F, and put
F2 = F. The result follows.
(b) (1) --+ (2): By Zorn's lemma, let II be an ideal in S maximal with h < a,
and put Fl = j,B. Dually, let 12 = Ill' and F2 a filter in S maximal with ,B < F 2.
(2) --+ (1): Let (Ij, Fj) E C* (j = 1,2) be two distinct ideal couples maximal with
respect to being properly contained in (I, F) such that h n h =I- 0 =I- Fl n F 2 . We
may assume that h C I. By Lemma 3.7, there is a E I with II < a. So, (la, FI) is
an ideal couple with (h,Fl ) ~ (la,FI) ~ (I, F). Hence (la,FI) = (I, F). Dually,
if F2 C F, we obtain that F = j,B for some ,B E S and thus the result. Therefore
suppose now that F2 = F = Fl. Since h n 12 =I- 0, by Lemma 3.7 we have either
h C h or h C h, which implies a contradiction to the maximality of (h, FI) or
(I2,F2).
The following shows that non-trivial intersections of ideal couples are ideal cou-
ples.

Lemma 3.12 Let (I, F), (I', F') E C' be ideal couples with InI' =I- 0 and FnF ' =I-
0. Then (I n I', F n F') is an ideal couple.

Proof. We only have to show that I n I' is upper directed and F n F' is lower
directed. Choose any a,,B E I n I'. There are 1 E I and {j E I' with {a,,B} :::; {I' {j}.
Let E E F n F'. Then {t, {j} :::; E. Since (S,:::;) is locally linear, we have 1 :::; {j or
{j :::; I, hence 1 E I n I' or {j E I n I'. Dually, F n F' is lower directed.

Let us call a normal subgroup N <l U a (maximal) ideal normal subgroup, if

N = NI,F for some (maximal) ideal couple (I, F) E C', respectively.
Observe that
Automorphism Groups of Generalized McLain Groups 111

Theorem 3.10 gave a group-theoretic characterization of the maximal ideal normal

subgroups of U. Now we use this result and Lemma 3.11 to derive a group-theoretic
characterization of the ideal normal subgroups of U, and of the normal subgroups
N!a,lfJ with a, (3 E S.

Corollary 3.13 Let N <l U be a normal subgroup.

(a) N is an ideal normal subgroup if and only if N = Nl n N2 for some maximal

ideal normal subgroups N 1 , N2 of u.
(b) The following are equivalent:

(1) N = N!a,lfJ for some a, (3 E S with a < (3.

(2) N is an ideal normal subgroup, and the collection of all ideal normal
subgroups properly contained in N contains at least two maximal elements
N 1 , N2 such that Nl n N2 i= 1.

Proof If (h, Ft), (h, F 2 ) E C* with h n 12 i= 0 and Fl n F2 i= 0, we have

(h n /2, Fl n F 2 ) E C* by Lemma 3.12 and NII,FI n N I2 ,F, = NhnI"FlnF,. Now
part (a) easily follows from Lemma 3.11(a), and part (b) from Lemma 3.11(b), using
also Lemma 3.3(b).
This result will be crucial in the following sections for the transfer of group-
isomorphisms to order-theoretic isomorphisms.

4 Baer-automorphisms of the McLain group

In this section we show that all Baer automorphisms of the McLain group G(R, S)
are locally inner, under suitable assumptions on Rand S. This also corrects a flaw
in the proof of the corresponding result for C(Fp, Q) in [18, Theorem 3].

Theorem 4.1 Let R be a ring with no zero-divisors i= 0 such that each automor-
phism of R is determined by its action on U(R), and let (S,::;) be an unbounded
locally linear connected poset. Then 0-( G) ~ Linn G.

Proof Let tJ E 0-( G). By Theorem 2.2, we may write tJ in the form tJ =
'1]*'0 ·.,p·T with either T = lor T induced by an order-anti-isomorphism T of (S,::;)
0-*
and a ring-anti-automorphism of R, and elements 0- E Aut S, 1] E Aut R, 0 E D
and .,p E Linn C. By assumption, (tJ) is subnormal in C{) = C Xl (tJ) and hence

[G, (tJ), ... , (tJ)] = 1 for some n EN.

--....--
n times

Suppose T i= 1, hence T i= 1. In order to find a contradiction, in a first step we

show that N := [G, (tJ)] contains elements 1 + eAI' for suitable A < /J in S. Because
112 Manfred Droste and Rudiger Gobel

of the commutator formula [x, yjZ = [x . z, y] . [z, y]-l, clearly N is normal in G.

We abbreviate a ' = aer (a E 5) and p := T2 E Aut 5. Let 1/ E 5. Since (5,::s:)
is unbounded, we can find elements A, fl. E 5 such that A < 1/ < fl., AT f. vper,
and I/T f. llper. Then a := vTer- 1 and (3 := ATer- 1 satisfy a < (3 by T f. 1 and
a f. fl.p, (3 f. vp. Now choose a E R. We compute

[1 + ae {3, 19]
a (1 - ae (3) . (1 + aeal {3' )'1 ·o¢ T
a

(1 - ae a (3)' + a ' e a l {31 )¢r

(1
(1 - aeo:(3) . (1 + a ' e",I{31 r . (1 + xcr l {3' r

where a ' E Rand Xcr l{31 E Xa l {3' is obtained as in the discussion of the locally
inner automorphisms of G following Theorem 2.2. Recall that X cr ' ,{3' contains only
the elements xo: ' ,{3' E E where each pair [8, c:] E [xo: 1,{3'] satisfies 8 ::s: a ' < (3' ::s: £
and either 8 < a ' or (3' < £. Also observe that the correspondence a --> a' is a
composition of two ring-automorphisms and multiplication with a unit of R, hence
it is a bijection of R which leaves U(R) invariant. Also, since (3' X:r> fl.T, for any =
bE R we have

[(1 + Xa l {31 r, 1 + bc IIP ,I'P] [1 + xa l {3', 1 - beI'T,IITY

((1- Xo:'{3' )(1 + beI'T,IIT)(l + xo:l{3')(l - beI'T,IIT)Y

(1 + beI'T,l/T . XIIT,.WY
1 + x~P,lIP' b' e IlP ,I'P
where X~p,IIP and b' , respectively, have entries obtained by transforming the ones
of XIIT,).,T, respectively, b with the ring-anti-automorphism inducing T, and X~p,IIP .
ellp,/'P E X).,p,I'P· Note that in contrast to the claim of [18, p. 275], this may be
non-trivial even in the case 5 = ((JJ, R = }~, as pointed out in Falkenberg [11, p.
77].
We have [1- aeo:{3, 1 + be llp ,I'P] = 1 since (3 f. 1/ P and a f. fl.p. Using the identities
[x. y, z] = [y, z] if [x, z] = 1 and [x. y, z] = [x, z]Y . [y, zJ, we obtain
[[1 + aCo:{3, 19], 1 + be llp ,I'P] = [(1- ae a (3)' (1 + a'eo:l{31 r .(1 + xa l {31 r, 1 + be vp ,I'P]
[( 1 + a ' e a , {3' r .(1 + Xo:l{3' r, 1 + be llp ,I'P]

[(1 + a ' e a l{3' r, 1 + be IlP ,I'P](1+x",,BI)' . [( 1 + xa l{3' r ,1 + beIlP,I'p]

[1- al/e)"p,lIp, 1 + bevp'l'p](1+X~P'PpJ . (1 + X~p,IIP . be IlP ,I'P)
(1 - al/be).,p,l'pp+x~p,pp) . (1 + x~p,llP . b'evp,/'P)

(1- al/be).,p,I'P)· (1 + x~p,vP' b' c lIp ,I'P)

1 - al/be).,p,I'P + X~p,IIP . b'e llp ,l'p,
The correspondence a ........ aI/ is again a bijection of R preservmg units. In
particular, if a is a unit and b = _(al/)-1, we have
Automorphism Groups of Generalized McLain Groups 113

Observe that
(1 + eAp,J.lp )(1+ye-",p) -- 1 + e y
Ap,J.lP - e. "",J.lp'
Moreover, conjugating this further with 1 + y' e,',Ap gives

1 + eAp,J.lp - y. e..."J.lP - y' . e...,',J.lp, if 1', 1" < Ap.

Together with (**), this shows that a suitable conjugate of[[l + ae"p, '19], 1 + bel/p,J.lp]
equals 1 + eAp,J.lp' Hence 1 + eAp,J.lp E N.
Now let N' := [N, ('19)] and choose Al < vI < 1'1 and ).,2 = A1PUT- 1 < V2 < 1'2
in S such that

(I) AiT i- ViPU, ViT i- J.LiPU for i = 1,2;

By (I) and the argument above, we have 1 + eAlP,J.llP E Nand 1 + eA,p,J.l2P E N.

Now define ex = AlP, f3 = J.L1P, A3 = f3uT- l = J.L1pu"f- 1, V3 = exUT- 1 = A1PUT- 1 =
A2, 1'3 = 1'2· Then A3 < V3 < 1'3 and A3T = J.L1PU i- A2PU = V3PU, V3T = A27 =
).,lPU i- J.L2PU = J.L3PU by (II). Also ex = AlP = V37U-l and f3 = J.L1P = A3TU - 1.
Hence, by the argument above, N' contains the element [[1 + e"p, '19], 1 + eI/ 3P ,J.l3P]
and so also 1 + e A3P ,/-'3P' Applying the same argument (n - 2) times again, we obtain
that [G, ('19), ... , ('19)] i- 1, a contradiction to (*). Thus T = l.
Next we show that U = l. Suppose U "I 1. There is A E S with A i- AU, and
we may assume that A < AU. For n E ~ let An = AUn . Then A = AO < Al <
A2 < .... Let t = 1 + eU l and in = t"n. Then we can write in in the form
in = 1 + bneAnAn+l + Xn with bn E Rand Xn E XAn,An+l' Also

with Yn E X An +l ,A n +2' Comparing this with the coefficient of eAn+1 ,A n +2 in in+1

shows that bn +1 is obtained by applying to bn two ring-automorphisms and mul-
tiplication with a unit of R. Inductively, we obtain bn i- 0 for each n E~. Now
consider the element [I, iI, ... , in]. By the special form of in, Lemma 3.2, and the
induction, we obtain that [I, iI, ... , in] E NA,A n and the coefficient of eA,A n+l is
bl · b2 · ... · bn i- O. Hence [I, iI, ... , in] "11 for each n E N. Thus the subgroup (i, '19) of
G" is not nilpotent and consequently G" is not locally nilpotent. However, since G
is a Baer-group (which coincides with its Baer radical) and '19 a Baer-automorphism,
G" is a Baer-group and hence locally nilpotent, a contradiction. Thus U = 1.
Finally we show that ",* {) = =
1. Since '19 ",* . {) .1/J, for any a E R, A < I' in Sand
x E XAJ.l with e = eA/-, we have (l+ae+x)" = l+a'e+x', and x' E XAJ.l' where the
correspondence a -> a' = ap' c is the composition of a ring-automorphism P = ",.1/J A
and multiplication with a unit c = cA/-' of R. In particular, a "I 0 implies a' "I O.
Thus [1 + ae + x, '19] = (1- ae - x) . (1 + a'e + x') = 1 + (a' - a)e - x + x'.
114 Manfred Droste and Rudiger Gobel

Suppose il is not locally inner. Then there is 0.0 E R with a~ t 0.0. Let
On = [0, (il), ... , (il)l. Since (il) is subnormal in O,J, we have Gn = 1 for some
'-.-'
n tinles
n E f;j. We now construct a sequence of elements Yi E G as follows. Put Yo =
1 + (o.~ - o.o)e. Suppose we have constructed Yi of the form Yi =
1 + (a' - a)e + x
for some a E R and x E X AI1 such that w := a' - a t o. Our first choice for
Yi+l is Yi+l := [Yi, ill = 1 + (w' - w)e + x', provided w :f:. w'. If w = w', then
we consider y(d) := 1 + w . de + Xd where d E [feR), Xd E X AI1 . Observe that
y(d) = yi for a suitable choice of a E Aut G and Xd E X AI1 . Then consider
[y(d), ill = 1 + ((w . d)' - w . d)e + x~. If (w . d)' t w . d for some d E U(R), then
wc choose such an clement d and call Yi+l := [y(d), ill. In any case Yi+l E Gf+l for
a suitable {3 E Aut G (composed of some of the a's). Since G n =
1, it follows that
the above construction process stops, say at step i + 1. So, for some a E R with
w = a' - a :f:. 0 and all dE U(R) we have
(1) w =w' and
=
(2) w . d (w . d),.
Next we determine the solutions of t.hese equations. We have
=
(w . d)' w . d = w' . d, and
w' = w P C, (w· d)' = wI' . d P . r: .
Hence
wI' .dP . C = wI' . C . d .
Since w :f:. 0 and R has no r,ero-divisors, it follows that d P = cdc 1, for all
dE [T(R). Sincc, by assumption on R, ring-automorphisms are determined uniquely
by their action on U(R), p is conjugation with c 1 . Hence, for allY l' E R we have
1" = 1'f'c = c· 1'. Now equation (1) implies 0 " - 2a' + a = 0, so (c 2 - 2c + 1)0. = O.
Since a :f:. 0 and R has no zero-divisors, this implies c = 1. Hence also p = 1 and in
particular as= aa, a cont.radiction.
As a consequence, we obtain the following chain of inclusions for the group 1+ pT
induced by the Jacobson radical, the Baer-radicals {3(1 + T*) and {3(Aut Gl, the
group 0-( G) of Baer-automorphisms and the locally inner automorphisms Linn G =
1 + T* of Aut G.

Corollary 4.2 Let R be a ring with no zero-divlsors t 0 such that each automor-
phism oj R is delermzned by its action on U (R). Let (S, S) be an unbounded locally
linear cOTlnected poset. Then

G c;:: 1 + pT c;:: ;3(1 + T*) c;:: j3(Aut G) c;:: o-(G) c;:: 1 + T* c;:: Aut G.

Proof. We noted before that clearly 0 = 1+E c;:: 1+pT. By Lemma 2.5, we have
1 + pT c;:: p( 1 + T*). The next inclusion is immediate by Corollary 2.4. The Baer
radical of Aut 0 is generated by cyclic subnormal subgroups (z) of Aut G, which
are clearly subnormal in (z, G) and hence contained in rr( 0). Finally, 0-( 0) c;:: 1 + T*
by Theorem 4.1 and Corollary 2.3.
Automorphism Groups of Generalized McLain Groups 115

This result will be crucial in the proof of our main theorem.

5 Proof of the main result

In this section we will show that if two McLain groups G( R, 5) and G( R', 5') have
isomorphic automorphism groups, then the McLain groups themselves are isomor-
phic. In fact, we show that the underlying posets and rings are either isomorphic
or anti-isomorphic. The argument will be based on Corollary 4.2 and the results of
section 3, in particular the purely group-theoretic characterization of ordered pairs
(a, j3) E 5 x S given in Theorem 3.10 and Corollary 3.13.
Therefore we make the following assumption throughout this section:

(5.0) R, R' are rings with no zero-divisors :f. 0 such that each automorphism
of R (resp. R') is determined by its action on U(R) (resp. U(R')), and
(5, :S), (5',:S) are unbounded locally linear connected posets.

Furthermore, let G = G(R, 5), G' == G(R', 5'), and A = Aut G, A' = Aut G'
and assume <p : A --+ A' is a fixed isomorphism. We will abbreviate No:(3 = N 1a ,T(3
for a, j3 E 5, a < j3. Note that by Lemma 3.3(b), No:(3 =
N,b iff a == / and j3 = 8.

Proposition 5.1 Let a, j3 E 5 with a < j3. Then No:(3<p == N a '(3' for some a', j3' E
5 with a' < j3'.

Proof. Clearly (j3A)<p == j3A', i.e. 'P maps the Baer-radical of Aut C onto that
of Aut C'. By Corollary 4.2, we have 1 + pT ~ j3A ~ 1 + T* ~ A, and by Theorem
3.10 and Corollary 3.13 the groups Na(3 can be characterized group-theoretically
within j3A. This implies the result.
Now put:l =
:1(5) = {[a,j3): a,j3 E 5, a < j3} where [a,j3) =
{O" E S:
a :S 0" :S j3}, an interval in 5, and similarly <1' = :1(8'). We define a mapplllg
zP : :I --+ :I' by
[a,j3]zp= [a',j3'] if Na(3<p= N a ,(3"
By Proposition 5.1 and Lemma 3.3(b), zp is well-defined and bijective.
We will show that this map is induced by an isomorphism or anti-isomorphism
between Sand 5'. For this, we need the following commutator formulas:

Lemma 5.2 Let a,j3",8 E 5 with a < j3 and / < 8.

(a) Let a = 1 + Eaieai(3,

i
E N a (3 and b = 1 + Ebje'JbJ E N,b. Then
j

i,j i.j
116 Manfred Droste and Rudiger Gobel

(b) [Nap, Nib] -I- 1 if and only if j3 ::; , or 8::; a.

Proof. (a) Easy, using a- 1 = 1 - LajC aiP ,' similarly b- 1 .

i
(b) If j3::; " we have 1 -I- 1 + Cab = [1 + Cap, 1 + epb] E [Nap, Nib] by (a). The
converse is also clear by ( a).

Lemma 5.3 (a) Lei I, J E J. If I ::; J then either lip ::; Jip or Jip ::; lip.
M oreove rIC;; J iff lip C;; Jip.

(b) Let a < j3 < , in 5 and [ai, j3'] = [a, j3]ip. Then there is " E 5' such that
either [j3,,]ip= [j3', ,'] and [a,,]ip= [a',,']
or [j3,,]ip= [r',a'] and [a, ,]ip = [r1,j3'].

(c) Let {a,8} < j3 in 5 and let [a l ,j3'] = [a,j3]ip and W,j3"] [8, j3]ip. Then
either j3' = j3" or 8' = a ' .

Proof. (a) Consider 1= [a,j3] and J = [r,8], first with j3::; ,. Let [a l ,j3'] = lip
and [r1,8 ' ] = lip. By Lemma 5.2, we have [Nap, Nib] -I- 1. Applying rp, we obtain
[NalfJ', N,lbl] -I- 1. Again by Lemma 5.2, this shows that either j3' ::; " or 8' ::; a ' .
lIence lip::; lip or lip::; lip. Now let I C;; 1. Then, ::; a < j3 ::; 8, so Nib C;; Nap.
Hence Ni' 61 C;; Ncr' fJ', showing " ::; a ' < j3' ::; 8' and thus lip C;; lip. The converse
follows dually.
(b) Let [a*,,'] = [a,,]ip. By (a), we have [a ' ,j3/] C;; [a*,,'], so a*::; a ' < j3'::; ,'.
Suppose we had a* < a' and j3' < ,'. Then by (a), J' = [j3,,]ipsatisfies l' C;; [a*,I']
and either JI::; [a l ,j3'] or [a l ,j3']::; 11. Thus either J' C;; [a*,a ' ] or J' C;; [j3',,'l-
In any case there is an interval contained in [a*, ,'] and disjoint (except, possibly,
for endpoints) to J' and [a l ,j3']. But there is no interval contained in [a,,] disjoint

,I,
(except for endpoints) to [j3, ,] and [a, j3], a contradiction to the properties of ip.
Hence either a* = a ' or j3' = but not both. If a* = ai, we obtain [j3, ,]ip C;;
[ai, ,'] and [a ' ,j3/]::; [j3"lip, so [j3"lip C;; [j3',,'J. By a similar argument as before,
it follows that this inclusion cannot be proper, proving our claim. If a* < a ' and
j3' = ,', similarly we get [j3, ,lip = [a*, a ' ], and our claim.
(c) Choose 1 E 5 with j3 < ,. We may assume, by (b), that [j3"lip = [j3', ,'] and
[a, ~(]ip = [ai, ,'] where a ' < j3' < " in 5' (in the other case argue dually). Again by
(b), we have either j3" = j3' or 8' = ,'. The latter case, however, is impossible since
then a' < j31 < " < j3", so [a, f3] ::; [8, j3] or [8, f31 ::; [a, j3] by (a), a contradiction.
As a consequence we obtain:

Corollary 5.4 Either for all I, J E J, I C;; J implies lip C;; Jip, or for all I, J E
:1, I C;; 1 implies lip C;; lip.

Proof. Let a < /3 < 1 < b in 5 and put 1 = [a,;3]' J = [f3, ,], K = [r,81.
Assume t.hat lip ::; Jip. Lemma 5.3 shows t.hat then Jip ::; Kip. Since 5 and 5'
 Automorphism Groups of Generalized McLajn Groups 117

are connected and unbounded, all intervals can be connected and we obtain that
I' ~ J' implies I'Tj5 ~ J'Tj5 for all I', J' E :J. If JTj5 ~ ITj5, we argue similarly.
We now define a mapping t.p' : S -+ S' induced by t.p as follows. Assume first that
ITj5 ~ JTj5 for all I, J E :J with I ~ J. Let a E S. Choose any {3 E S with a < {3
and let [a',{3'] = [a, {3]Tj5. We put at.p' = a'. By Lemma 5.3(c) t.p' is well-defined. If
=
, E S with {3 < " then [{3, ,]Tj5 [{3', ,'] by (5.3)(b), so {3t.p' =
{3' > a' = at.p' and
t.p' is order-preserving. Since t.p is an isomorphism, t.p' is an order-isomorphism.
Now assume, by Corollary 5.4, the alternative case that JTj5 ~ ITj5 for all I, J E :J
with I ~ J. Let a E S. Choose f3 E S with 0'< {3 and let [a',f3'] = [a,{3]Tj5. We put
=
at.p' {3'. A similar argument shows that t.p' : S - S' is an order-anti-isomorphism.
Hence we have:

Proposition 5.5 Under the assumption (5.0), the induced mapping t.p' : S -+ S' is
either an order-isomorphism or an order-anti-isomorphism satisfying, correspond-
ingly, Na(Jt.p Na<pI,(J<p1 for all (\' < f3 in S, or Na(Jt.p = N(J<pI,a<p1 for all (\' <
f3 in S.

Subsequently, we will write (\" = at.p', {3' = {3t.p' for a < {3 in S. We let Ya(J
comprise all elements y E T* such that for any (" 6) E [y] we have , ~ (\' < {3 ~ 6
and either, < a or {3 < 6. (Thus Ya(J n E = Xa(J defined before.) Next we turn
to the proof that Rand R' are isomorphic or anti-isomorphic. Let a < {3 in S. We
define a mapping t.pa(J : R -+ R' as follows. For a E R, we have g = 1 + aea(J E N a(J.
First let t.p' be order-preserving. Then gt.p E N alf3l. So, gt.p = 1 + a'eal(J' + y
where y E Yal(J'. Then put at.pa(J = a'. Secondly, if t.p' is order-reversing, we have
gt.p E N(J',a " Hence gt.p = 1 + a'e(J'al + y where y E Y(J'a" Again put at.pa(J = a'.

Proposition 5.6 (a) For any(\'< {3 in S, t.pa(J : R+ -+ R' + is an isomorphism.

(b) For any a < {3 < , in S and a, bE R, the following equation holds:

(at.pa(J)· (bt.p(J-y) (ab )t.pa-y, if t.p' is order-preserving, and

(bt.p(J-y)' (at.pa(J) -( ab )t.pa-y, if t.p' is order-reversing.

Proof. (a) For any a, bE R we have (1 + ae a(3) . (1 + bea(J) = 1 + (a + b)e a f3.

Applying t.p and computing the product in 0', we get (a + b)t.pa(J = at.pa(J + bt.pa{3'
We now show that t.pa(J(t.p-l )al(J' = id R . Let a E Rand g = 1 + ae a f3. Then, as
noted above, gt.p = 1 + (at.pa(J)e a1f31 + y' where y' E Y~/(J" Since eal(J1 . y = 0, we
have gt.p = (1 + (at.pa(J)eal(J') . (1 + y'). Applying t.p-l to this equation, we obtain
9 = (1+at.pa(J(t.p-l )al(J'e a(J+y)·(1+y')t.p-l where y E Ya(J and (1+y')t.p-l E 1+Yaf3 .
So at.paf3( t.p;'/~/) = a. It follows that t.pa(J is an isomorphism.
(b) We have [1 + ae a {3, 1 + be(J-y] = 1 + abea-y by Lemma 5.2(a). First let t.p'
be order-preserving. Applying t.p, we get (1 + aea(J)t.p = 1 + (at.pa(J)eal(J1 + y' and
(1 +be(J-y)t.p = 1+ (bt.p(J-y )e(JI-y1 +z'. So, computing the commutator in 0' with Lemma
5.2( a), we have (1 + abea-y)t.p = [(1 + aea(J )t.p, (1 + be(J-y )t.p] = 1 + (at.pa(J )(bt.p{3-y )eal-yl +
118 Manfred Droste and Rudiger Gobel

x' where x' = (a<paj3e a 1j31) . Zl + y' . (b<pj3-y)ef3'-y1 + y' . Zl and x, E Y~If3'. Hence
(a<paj3)(b<pj3I') = (ab)<Pa-y. If <p' is order-reversing, perform a similar calculation.
For any a: < (3 in 8, we put ca j3 = l<paj3 E R'.
Proposition 5.7 For any a: < (3 in 5, ca j3 is a unit of R'. For a: < (3 in 8, define
1/Jaj3 : R --+ R' by a1/Jaj3 = (a:<P"'j3) . c:~ for all a: E R. If <p' is order-preserving,
then 1/Jaj3 is a r·ing-isomorphism. If <p' is order-reversing, then 1/Jaj3 is a ring-anti-
isomorphism.

Proof. Let b < a: < (3 < 'Y and assume <p' is order-preserving. By Proposition
5.6( a), <Pa-y and <Pbj3 are bijective, so there are y', y E R with Y<P"'-Y = 1 = y' <P6j3.
Then by (5.6)(b) we obtain 1 = (1 . Y)<Pa-y = ca j3 . (Y<Pj3-y) and 1 = (Y' . 1)<Pbj3 =
(y' <Poa) . caj3· Clearly, y' <Poa =(y' <Po",) . Caj3 . (Y<Pj3-y)= y<Pj3-y. So, caj3 is a unit in
R'. By (a) and Proposition 5.6(a), 1/J",-y : R+ --+ R+ is an isomorphism. Now let
a, bE R. Then using the functional equation (5.6)(b),

(a· 1)<Pal' . c~~ . (1 . b)<pa-y . c~~

(a<paj3)· cj31' . c~~ . caj3· (b<pj3-y). c~~
(ab)<Pa-y . c~~
(ab)1/Ja-y.

So, 1/Ja-y is a ring-isomorphism. If <p' is order-reversing, a similar computation ap-

plies.
Now we can summarize our results:

Main Theorem 5.8 Let R, R' be rings with no zero-divisors f. 0 such that each
automorphism of R (resp. R') is determined by its action on U(R) (resp. U(R')),
and let (8, ::;), (8' ,::;) be unbounded locally linear connected posets. Let Aut G(R, 5)
and Aut G(R', 8 ' ) be isomorphic. Then G(R, 8) and G(R', 8' ) are isomorphic.

Proof. By Propositions 5.5 and 5.7, we obtain either a pair of isomorphisms

<p' : 5 --+ 8 ' , 1// : R --+ R' , or a pair of such anti-isomorphisms. Similarly as
described in section 2, such pairs induce an isomorphism between G(R,8) and
G(RI,5' ).

References
[1] A. L. S. Corner, B. Goldsmith, Isomorphic automorphism groups of torsion-free
p-adic modules, submitted.
[2] J. Dieudonne, La geometrie des groupes classiques, Vol. 5, Ergebnisse der Math-
ematik, Springer, Berlin, Heidelberg, New York, 1991.
Automorphism Groups of Generalized McLain Groups 119

[3] .J. Dieudonne, On the automorphisms of classical groups, Memoirs Amer. Math.
Soc. 2(1950).
[4] M. Droste, Structure of partially ordered sets with transitive automorphism
groups, Memoirs Amer. Math. Soc. 334(1985).

[5] M. Droste and R. Gobel, McLain groups over arbitrary rings and orderings,
Math. Proc. Camb. Phil. Soc. 117(1995), 439~467.

[6] M. Droste, C. Holland, and D.Macpherson, Automorphism groups of infinite

semilinear orders (I), (II), Proc. London Math. Soc. (3) 58(1989), 454~478
and 479~494.

[7] M. Dugas and R. Gobel, Automorphism groups of torsion-free nilpotent groups

of class two, Trans. American Math. Soc. 332(1992), 633~646.

[8] M. Dugas and R. Gobel, All infinite groups are Galois groups over any fields,
Trans. American Math. Soc. 304(1987), 355-384.
[9) M. Dugas and R. Gobel, On locally finite p-groups and a problem of Philip
Hall's, J. Algebra 159(1993), 115~138.

[10] M. Dugas and R. Gobel, Automorphism groups of fields, Manuscripta Mathe-

mati cae 85(1994), 227~242.
[11] H. Falkenberg, Die McLain-Gruppen iiber allgemeinen Ketten und ihre Auto-
morphismengruppen, Diplomarbeit, Universitat Essen, 1993.

[12] R. Gobel, Wie weit sind Moduln vom Satz von Krull-Remark-Schmidt ent-
fernt?, Jahrcsber. DMV 88(1986), 11-49.

[13] P. Hall, Wreath powers and characteristically simple groups, Math. Proc. Camb.
Phil. Soc. 58(1962), 170~184.
[14) H. Leptin, Abelsche p-Gruppen und ihre Automorphismengruppen, Math.
Zeitschr. 73(1960),235-253.
[15) W. Liebert, Isomorphic automorphism groups of primary abelian p-groups, in
Abelian Group Theory, (R. Gobel, E. A. Walker, ed.), Gordon and Breach,
London, 1987,9-13.
[16) D. H. McLain, A characteristically-simple group, Math. Proc. Camb. Phil. Soc.
50(1954),641-642.
[17] D. J. S. Robinson, A Course in the Theory of Groups, Vol. 80, Springer Grad-
uate Texts, 1980.
[18) .J. E. Roseblade, The automorphism group of McLain's characteristically simple
group, Math. Zeitschr. 82(1963),267-282.
120 Manfred Droste and Rudiger Gobel

[19] J .S. Wilson, Groups with many characteristically simple subgroups, Math. Proc.
Camb. Phil. Soc. 86(1979), 193-197.
 Locally Moving Groups and Reconstruction Problems

Matatyahu Rubin*

Department of Mathematics
Ben Gurion University of the Negev
Beer Sheva
Israel

Abstract
Let B be a complete atomless Boolean algebra. For 9 E Aut(B) let
var(g) d~ I)a E B I g(a) . a = a}. Let G ~ Aut(B). We say that Gis
a locally moving subgroup of Aut(B) if {var(g) I 9 E G} is dense in B. We
show (Corollary 1.4) that for every complete atomless Boolean algebra Band
a locally moving subgroup G of Aut(B), the group G determines B up to
isomorphism. We then use this theorem to prove that certain mathematical
structures are determined by their automorphism group. In Section 3 we show
such a reconstruction theorem for a certain class of locally com pact topological
spaces. Then we prove a reconstruction theorem for subspaces of the foliated
torus. In Section 4 we present a reconstruction theorem for 2-homogeneous
linear orders and circular orders.

1 Introduction
The goal of this paper is to prove a new theorem about locally moving groups, and
demonstrate the role of such groups in certain reconstruction problems.
A reconstruction theorem is a result of the following form. A class f{ of mathe-
matical structures is given, and it is proved that for every M, N E J{ and isomor-
phism i.p between Aut(M) and Aut(N) there is an isomorphism T between M and
N such that for every f E Aut(M), i.p(J) = TO f 0 T- 1 .
We next present the notations needed for the definition of a locally moving
group. A more detailed treatment of this notion and of some prerequisites will be
given in Section 2.
'This work was prepared in 1994-95 while the author was a visitor in Bowling Green State
University, Bowling Green, Ohio and in the University of Colorado, Boulder, Colorado. The author
is most grateful to both of these institutions.

121
W. C. Holland (ed.), Ordered Groups and Infinite Permutation Groups, 121-157.
© 1996 Kluwer Academic Publishers.
122 Matatyahu Rubin

Notation and Definitions 1.1 Let B be a Boolean algebra (BA).

(a) The operations, units and partial ordering of B are denoted respectively by
+B, .B, _B, OB, IB and ~B. If A is a subset of B, then 'LB A and TIB A denote
respectively the supremum and infimum of A in B whenever they exist.
The superscript B is omitted whenever no confusion may arise.
(b) B is atomless, if (B - {O}, ~) has no minimal elements.
(c) B is complete, if (B, ~) is a complete lattice. That is, every subset of B has
an infimum and a supremum.
(d) Let a E B, then Bla d~ {b E Bib ~ a}.
(e) Let Aut(B) denote the automorphism group of B. Let 9 E Aut(B) and
a E B. Then gla d~ g[(Bla).
(f) Suppose that B is a complete BA. Let 9 E Aut(B), then fix(g) d~ 'L{a E
B I gla = Jd} and var(g) d~ -fix(g).
(g) A subset D of B is dense in B, if for every bE B - {OJ there is dE D - {OJ
such that d ~ b.
Definition 1.2 Let B be a complete atom less BA and 0 ~ Aut(B). 0 is a locally
moving subgroup of Aut (B) if {var(g) I 9 E O} is dense in B. The pair (B, 0) is
then called a local movement system.
A group H is a locally moving group if for some complete atomless BA B, H is
isomorphic to a locally moving subgroup of Aut(B).
Theorelll 1.3 (The Expressibility Theorem)
There are first order formulas ipEq(X, y), ip~(x, y) and ipAp(X, y, z) in the language
of groups such that for every local movement system (B, 0), the following holds.
For every f, g, hE 0: 0 F ipEq[f, g] iff varU) == var(g), 0 F ip~ [f, g] iff varU) ~
var(g) and 0 F ip Ap [f, g, h] iff f( var(g)) = var( h).
The Expressibility Theorem can be used in answering two closely related types
of questions. The first one concerns reconstruction questions, and the second one
is about elementary equivalence of automorphism groups.
The following corollary is a reconstruction theorem that follows quite easily from
the Expressibility Theorem. This corollary is later used as a first step in the proof
of other reconstruction theorems.
Corollary 1.4 (The Reconstruction Theorem for Local Movement Systems)
For i =
1,2 let (Bi, G i ) be a local movement system. Suppose that ip is an isomor-
phism between G 1 and G 2 . Then there is an isomorphism 7 between B1 and B2
such that for every f E G 1, ip(!) == TO f 0 7- 1 .
We shall discuss the elementary equivalence questions later. That discussion
requires some more definitions and notation. Next we mention an isolated result (it
has so far no continuations), that seems to be of interest. It also follows relatively
easily from the Expressibility Theorem.
 Locally Moving Groups and Reconstruction Problems 123

Corollary 1.5 There is a first order sentence t.pLM in the language of groups such
that for every group H: H F= <P LM iff H is locally moving.

Weaker variants of the expressibilty theorem appear in [12, 13, 14, 15]. Those
variants include more assumptions on G on top of its "local movingness". These
weaker theorems sufficed for the various applications that were proved there. How-
ever, the present expressibility theorem has new applications that could not be
proved using the weaker variants.
In the remaining part of the Introduction we describe some settings in which lo-
cally moving groups occur. In these settings the Expressibility Theorem can be used
in the proof of reconstruction and elementary equivalence results. In the last part
of the introduction we discuss the topic of elementary equivalence of automorphism
groups.
The article is organized in such a way that the reader can skip the definitions
and theorems concerning elementary equivalence, and the remaining part of the
article which concerns only with reconstruction problems remains self-contained.
The proof of the Expressibility Theorem and its immediate corollaries will be
presented in Section 2.
Section 3 describes two applications in topology. The first one concerning locally
compact spaces is old. The second one on foliations of the torus is new.
Section 4 describes an application for linear and circular orders. All the results
in that section are old. However, we believe that the unified treatment of several
different structures is new, and adds something to the understanding of the phe-
nomenon of reconstructibility of such structures. Section 4 does not use the results
of Section 3. Once the part of the Introduction about homeomorphism groups has
been read, Section 4 can be read independently of Section 3.
We now describe a number of settings in which locally moving groups occur. In
these settings locally moving groups were used in order to prove various reconstruc-
tion theorems.
In what follows, it will be convenient to use the notion of a "faithful class" .

Definition 1.6 (a) Let K be a class of pairs of the form (M, G), where
M is structure and G:S Aut(M). The class K is faithful, if for every
(Ml' G 1 ), (M2' G 2 ) E K and an isomorphism t.p : G 1 == G 2 there is an isomorphism
T : Ml == M2 such that T induces <po
(b) A class K of structures is said to be faithful if {(M, Aut(M)) 1M E K} IS
faithful.

Homeomorphism groups of topological spaces

Let H(X) denote the group of self-homeomorphisms of a topological space X.
We wish to show that under appropriate assumptions on X, X can be recovered
from H(X). Indeed, we shall be able to recover X from various subgroups of H(X).
We associate with every topological space a complete Boolean algebra
124 Matatyahu Rubin

Definition 1.7 Let X be a topological space. (a) For U ~ X, int(U) and cl(U)
denote respectively the interior and closure of U.
A subset U of X is regular open in X, if U = int(cl(U». Let Ro(X) denote the
set of regular open subsets of X.
(b) The poset (Ro (X),~) is the partial ordering of a complete BA. The opera-
tions of this BA are U . V = un V, U + V = int( cl(U U V» and", U = int(X - U).
We thus regard Ro(X) as a Boolean algebra.
(c) For 9 E H(X) let gRo : Ro{X) -> Ro{X) be defined as follows: gRo (U) =
{g(x) I x E U}. For 0 ~ H{X) let ORo = {gRo I 9 EO}.

We next describe the connection between locally moving groups and the action
of H(X) on X.

Proposition 1.8 Let X be a HausdorfJ space.

(a) The function 9 1-+ gRo is an embedding of H(X) into Aut(Ro(X».
We shall later regard H(X) as a subgroup of Aut(Ro(X».
(b) For g E H(X), fix(gRo) = int({x E X I x =g(x)}) and var(gRo)
int{cl({x E X I x of g(x)}».
(c) Let 0 be a subgroup of H(X); then ORo is a locally moving subgroup of
Aut(Ro(X» iff

(*) for every nonempty open subset U of X there is 9 E 0 - {Jd} such that
gl(X - U) =!d.
Note that if X is an open subset of IR n, then H(X) and many of its subgroups
have property (*) from the above proposition. This is also true for many other
topological spaces.
In Theorem 3.1 we shall use this fact in order to conclude that the class of open
subsets of IR n regarded as topological spaces is faithful. In fact, Theorem 3.1 is
considerably stronger than this. The faithfulness of Euclidean manifolds is due to
Whittaker [20].
Reconstruction theorems in topology have been proved by quite a number of
people. The works of Whittaker [20], Ling [8] and the author [12] can be mentioned.
This should not be regarded however as a description of the history of this topic.
As an example of such a result we mention here a theorem from [16].

TheoreIll 1.9 Let LIP (X) denote the group of bi-Lipschitz self-homeomorphisms
of a metric space X. Let]{ {(X, 0) I X is an open subset of a
Banach space, and LJP(X) ::; 0::; H(X)}. Then]{ is faithful.

AutoIllorphisIll groups of linearly ordered sets

Definition 1.10 A linear ordering (L, <) is 2-homogeneous if ILl> 2 and for every
Xl < X2 and Yl < Y2 in L there is 9 E Aut«(L, <) such that for i = 1,2, g(Xi) = Yi.
Locally Moving Groups and Reconstruction Problems 125

We regard linear orderings as topological spaces equipped with their order topol-
ogy. Then Aut«(L,<)) is a subgroup of H(L). It is easy to see that if (L,<) is
2-homogeneous, then Aut«(L, <}) is a locally moving subgroup of Aut(Ro(L)).
It will follow easily from this fact that up to the reversal of its ordering, the
Dedekind completion of a 2-homogeneous linear ordering is determined by its au-
tomorphism group.
The Reconstruction problem of 2-homogeneous linear orders has a long history.
Holland [6J proved the reconstructibility of the completion of a chain from its au-
tomorphism group regarded as lattice ordered group. Then Rabinovich [l1J 1975
and McCleary [10] 1978 independently proved the reconstructibility of the comple-
tion of a chain from the automorphism group regarded only as a group. McCleary
also showed there that the chain is determined by rich enough subgroups of its full
automorphism group.
The question of distinguishing between automorphism groups of chains by first
order sentences was dealt with by Gurevich and Holland in [4]. Among other things
they proved that there are first order sentences in the language of group theory
that distinguish Aut(JR) and Aut( Q) from the automorphism groups of all other
2-homogeneous chains (the irrationals excluded).
In Section 4 we also deal with circular orders. These are structures which are
obtained by turning linear orders into circles. The exact definition appears in 4.1.
I do not know whether the reconstruction result for 2-homogeneous circular orders
has ever been explicitly published. In any case, it is a special case of [12] Theorem
3.5. It may have been known before.
Recently McCleary and the author strengthened the known reconstruction re-
sults on orders and circular orders. We proved that the Dedekind completion of
such structures is determined up to order or orientation reversal by any subgroup
of the automorphism group of the structure which is approximately 2-transitive
and locally moving. The reconstruction uses only first order formulas. These re-
sults will appear elsewhere. (A group of automorphisms of a chain or a circular
order L is approximately 2-transitive if for every distinct Xl, X2 ELand open inter-
vals h, h ~ L: if there are Yl E h and there Y2 E 12 such that the map Xi ~ Yi,
i= 1,2 is a partial isomorphism, then there is 9 E G such that for every i = 1,2,
9(Xi) E Ii') A variant of the notion of approximate transitivity was defined by Bieri
and Strebel in [2]. There, they reconstruct the real line from subgroups that have
stronger approximate transitivity properties.

Automorphism groups of trees

There are different mathematical objects that are called trees. Here the notion
of a tree stands for a certain type of a partially ordered sets. However, graphs which
are trees and structures with a betweenness relation which are trees also have locally
moving automorphism groups if they are sufficiently homogeneous.
126 Matatyahu Rubin

Definition 1.11 (a) Let M = (T, <) be a partially ordered set. M is a tree if for
every sET the set {t E Tit < s} is linearly ordered by <.
(b) Let M = (T, <) be a tree. M is an ever-splitting tree, if for every sET
there are tl, t2 > s such that t1 and t2 are incomparable.

In order to keep the presentation simple, we shall deal only with ever-splitting
trees. We shall associate with every such tree a complete atomless BA.

Definition 1.12 Let M = (T, <) be an ever-splitting tree. For sET let M~s =
{t E Tit ~ s}. Let B1 (M) be the subalgebra of the power set of T generated
by {M~S Is E T}. Let heM) be the ideal of B 1(M) consisting of all finite sets
belonging to B1(M), and let B 2 (M) = B 1(M)/h(M). Let B(M) be the completion
of B 2 (M).

It is obvious that if M is an ever-splitting tree, then Aut(M) can be regarded

as a subgroup of Aut(B(M)). We shall next find a condition which assures that
Aut(M) is a locally moving subgroup of Aut(B(M)).

Proposition 1.13 Let M = (T, <) be an ever-splitting tree.

(aJ B(M) is atom/ess.
(bJ Aut(M) is a locally moving subgroup of Aut(B(M)) iff

(**) for every sET there is g E Aut(M) and t > s such that g(s) = sand
get) f. t.
Note that the binary tree, i.e., the tree of all finite {O, l}-sequences has property
(**) from the above.
The reconstruction problem for trees is dealt with extensively in [15]. There, a
nearly complete answer is given to the question of when two trees have isomorphic
automorphism groups. The analysis of the linear case is included there as a special
case.

Automorphism groups of incomplete Boolean algebras

If B is a BA, then every automorphism of B can be extended in a unique way
to an automorphism of the completion of B. Let B denote the completion of B.
We may thus regard Aut(B) as a subgroup of Aut(B).

Proposition 1.14 Let B be an atomless BA. Then Aut(B) is a locally moving

subgroup of Aut (B) iff for every a E B - {O} there are distinct bI , b2 ::; a such that
Blb 1 ~ Blb 2 •

The reconstruction question for incomplete Boolean algebras is dealt with in

[13, 14]. The results in [14] precede those of [13], but they already include strong
reconstruction theorems for homogeneous Boolean algebras. The results in Theorem
Locally Moving Groups and Reconstruction Problems 127

4.5 of [13] are much stronger. They do not assume the homogeneity of B, but rather
a much weaker assumption. Let us state here a special case of Theorem 4.5 of [13].
This will indicate the type of assumptions that suffice for a reconstruction theorem
in this context.

Definition 1.15 (a) Let B be a BA and G ~ Aut(B). G is restriction-closed (R-

closed) if (i) for every 9 E G and a E B, if g(a) = a, then the automorphism of
B extending 9 raUld r - a belongs to B; and (ii) for every 9 E G and a E B, if
g(a) . a = 0, then the automorphism of B extending gla U g-llg(a) U Idl- (a + g(a))
belongs to B.
(b) For a BA B let Ult (B) denote the set of ultrafilters of B. Note that Aut (B)
acts on Ult(B).

Theorem 1.16 Let f{ be the class of all pairs (B, G) such that:
(i) B is an atomiess RA, and G ~ Aut(B); (ii) G is a locally moving subgroup of
Aut(B); and (iii) for every p E Ult(B), 3 ~ I{g(p) I 9 E G}I # 6.
Then f{ is faithful.

Groups of measure preserving automorphisms

Definition 1.17 (a) Let B be an atomless u-complete BA, (that is, a Boolean
algebra in which every countable subset has a supremum and an infimum). A
function J.L from B to the non-negative extended reals is a strictly positive measure
on B, if the following holds.
= =
(1) For every a E B, J.L(a) 0 iff a O.
(2) For every countable subset A of B of pairwise disjoint elements, J.L("L. A) =
"L.{J.L(a) I a E A}.
The pair (B, J.L) is then called a measure algebra.
(b) Let (B, J.L) be a measure algebra.
Aut«(B,J.L)) d~ {g E Aut(B) I for every a E B, J.L(g(a)) = J.L(a)}.
Measure algebras can be obtained by dividing the u-algebra of a measure space
by its ideal of measure zero sets.
It follows from a theorem of Maharam [9] that for every measure algebra (B, J.L),
Aut«(B, J.L)) is a locally moving subgroup of Aut(B)}.
Section 6 of [13] describes many results, and several open questions in this topic.

Quotient groups acting on quotient algebras

Recently a new type of reconstruction theorem was considered. We demonstrate
this type of question by an example.
128 Matatyahu Rubin

Question 1.1 Let BH(JR") denote the group ofself-homeomorphismsofJR" whose

support is bounded, that is, BH(JR") d~ {g E H(JR") I{x I x:f- g(x)} is bounded}.
Let QBH(JR") = H(JR")/ BH(JR").
Is it true that for every automorphism cp of QBH(JR") there is g E H(JR") such
that g . BH(JR") induces cp?

In this setting the group QBH(JR") acts in a locally moving fashion on the quo-
tient of Ro (JR n) over the ideal of bounded regular open sets. So the Reconstruction
Theorem for Local Movement Systems can be invoked.
This situation occurs also in other settings, e.g. the setting of linear orderings.
However, only a few final results are presently known.
The first results in this direction are theorems on the recovery of the quotients of
the power set of K, P(K), from the corresponding quotients of the symmetric group
Sym (K). Two such theorems concerning reconstruction and elementary equivalence
appear in (14] Theorems 4.2 and 4.3. The full answer to the elementary equivalence
question of quotients of the symmetric group was recently found by Shelah. It
proves the cases that were not covered in (14] Theorem 4.3. It will appear in (17].
A complete reconstruction result for the quotient of P( w) is due to Alperin,
Covington, and Macpherson (1] in this volume.
Another reconstruction result from quotients is due to Giraudet and Truss [5]. It
concerns the quotient groups of Aut( (JR, <)) and Aut( (Q , <)). The exact statement
of this theorem appears in the article of Truss [19] in this volume.
The author has recently generalized Giraudet and Truss' result to the class of 2-
homogeneous linear orderings whose cofinality is No. However, the general question
for 2-homogeneous linear orderings is still open.

Elementary equivalence of automorphism groups

The Expressibility Theorem can be used in proving that certain automorphism
groups are not elementarily equivalent. We start with some definitions.

Definition 1.18 We use the terminology of model theory. We let M, N denote

structures. IMI denotes the universe of M. If R is a relation or a function sym-
bol in the language of M, then its interpretation in M is denoted by RM. If
cp( Xl, ... , xn) is a formula in the language of M whose free variables are among
Xl,···, x n , and al,···, an E IMI, then M 1= cp[al, ... , a'll means that cp holds in M
in the assignment Xl f-+ ai, ... , Xn f-+ an.
(a) Let M and N be structures in the same language L. We say that M and N
are elementarily equivalent (M == N) if for every first order sentence cp in L: M 1= cp
iff N 1= cp.
(b) Let M be a structure in a language Land G be a group of permutations of
IMI. (G will usually be a group of automorphisms of M.) The structure N defined
below is called the action structure of M and G. INI = IMI u G. The language of N
is L U {P, 0, Ap}, where P is a unary predicate, and 0 and Ap are binary function
Locally Moving Groups and Reconstruction Problems 129

symbols. For every relation symbol R E L, RN = RM. If F E L is a k-place

function symbols, then FNtlMlk = FM; we extend this to INlk in some trivial way;
a
say, for every = (a1, ... ,ak) E INlk - IMlk, FN (ii) = al. The interpretation of
the new symbols is as follows. pN = IMI. oN is the composition operation on G;
and oN is defined in some trivial way on IN 12 - G 2 . ApN is the application function
from G x IMI to IMI. That is, if 9 E G and a E IMI, then ApN (g, a) = g(a). ApN
is defined in some trivial way on INI2 - 0 x IMI. We denote N by Act(M, 0).
(c) For a local movement system (B, 0), let
Var(B,O) d;J ({ var(g) I9 E O}, ~B). We use the notation Var(B, G) to also denote
just the set {var(g) I 9 E G}.

Part (a) of the following corollary demonstrates the way that the Expressibility
Theorem can be used in showing that certain automorphism groups are not ele-
mentarilyequivalent. However, in applications, it is instead Part (b) which is being
used. It is also used in the proof of Part (a).

Corollary 1.19 (aj For I = 1,2 let (Bl' G 1) be local movement systems. Then if
C 1 == O 2 , then Act( Var(B l , Cd, Cd == Act( Var(B 2 , C 2 ), O 2 ).
(b j For every first order formula <pC Xl, ... , Xk, Yl, ... , Ym) in the language of
Act( Var(B, C), C) there is a first order formula <P*(X1' ... ' Xk, Yl, ... , Ym) in the
language of groups such that for every local system (B, C), h E C k
and g E cm, Act(Var(B,C),C) F <p[var(hd,···,var(hk),gl, ... ,gm] iff 0 F
<p*[h,ff].

The fact that (a) follows from (b) is trivial. It is quite easy to see how Part (b)
follows from the Expressibility Theorem. So we shall not return to the proof of this
corollary.
Theorem 4.6(b) is an example of the use of the above corollary in answering
questions concerning the elementary equivalence of automorphism groups of linear
orders and related structures.

2 Locally moving groups

In order to make this section self-contained, we repeat some of the notions that
were defined telegraphically in the introduction. Locally moving groups are defined
in 2.4, and the Expressibility Theorem is restated in 2.5.

Notation 2.1 Let B be a BA.

( a) Let a E B. B [a d;J {b E Bib ~ a}.
(b) Let 9 E Aut(B) and a E B; then g[a d;J g[(B[a).
(c) Suppose that B is complete. Let 9 E Aut(B);
then fix (g) d;J L:{a E B I g[a = Id}.
130 Matatyahu Rubin

= -fix(g ) .
(d) var(g) de!
(e) For a group G and g, hE G let gh = hgh- I and [g, h] = ghg-Ih- I .

Proposition 2.2 Let B be a BA, A ~ Band b E B. Then the following are

equivalent:
(1) For every bi E BIb - {OJ there ~s a E A such that bi . a f= O.
(2) For every upper bound a of A, b::; a.

Proof: (1) => (2). Let a be an upper bound of A. Let c = b - a. So c is disjoint

from a. So it is disjoint from every a' ::; a. So for every a' E A, c· a' = O. Since
c ::; b, c = O. SO b::; a.
(2) => (1). Let b' E BIb - {OJ, and suppose by contradiction that for every
a E A, b' . a = O. Hence -b' is an upper bound of A. So b ::; -b' , a contradiction.
o

Proposition 2.3 (a) Let B be aBA, 9 E Aut(B), and a E B be such that g(a) f= a.
Then there is 0 < b::; a such that g(b) . b = O.
(b) Let B be a complete BA and 9 E Aut(B). Then
var(g) = L{a E B I a· g(a) = OJ.

Proof: (a) If a - g(a) f= 0, then a - g(a) is the required b. Otherwise g(a) > a.
= =
Let c g(a) - a and b g-l(c). Then b is as required.
(b) Let Band 9 be as in part (b). Let A = =
{a E B Ig(a)·a O} and ao = LA.
For every a E A, a . fix(g) = O. So ao . fix (g) = O. So ao ::; -fix (g) = var(g). We
show that var(g) ::; ao. Let 0 f= b ::; var(g). So glb f= fd. Let c E BIb be such that
g(c) f= c. Hence there is 0 f= a ::; c such that g(a) . a = O. So a E A. By Proposition
2.2, var(g) ::; ao. 0

Definition 2.4 Let B be an atom less complete BA and G ::; Aut(B). G is a locally
moving subgroup of Aut(B), if for every a E B - {OJ there is 9 E G - {fd} such
that var(g) ::; a. The pair (B, G) is called a local movement system.
We say that a group G is a locally moving group, if there is a complete atom less
BA B such that G is isomorphic to a locally moving subgroup of Aut(B).

Theorem 2.5 (The Expressibility Theorem)

There are first order formulas <f'Eq(X, y), <f'~(x, y) and <f'Ap(X, y, z) in the language of
groups such that for every complete atomless BA B and locally moving subgroup G ::;
Aut (B) the following holds. For every f, 9 E G: G F <f' Eq [I, g] iff var(J) = var(g);
G F <f'<[f,g] iffvar(J)::; var(g); and G F <f'Ap[J,g,h] iff f(var(g» = var(h).

The proof of Theorem 2.5 will be completed at the end of this section.

Definition 2.6 Let B be a BA and D ~ B. D is dense in B if for every b E B- {OJ

there is dE D - {OJ such that d::; b.
 Locally Moving Groups and Reconstruction Problems 131

Remark 2.7 The definition of a locally moving subgroup of Aut(B) can be re-
phrased as follows. G S Aut(B) is locally moving if {var(g) I 9 E G} is dense in
B.

Proposition 2.8 (a) If D is dense in B, then for every a E B,

a=I:{dEDldsa}.
(b) For i = 1,2 let Bi be a complete BA and Di be a dense subset of B i . Let
a : (D1' s) ~ (D2, S)· Then a can be extended to an isomorphism between Bl and
B 2 , and this extension is unique.

Proof: The proposition is well-known, and its proof is easy. 0

Notation 2.9 Let B be a Boolean algebra and G S Aut(B).

M(B, G) d~ (B, G; S, 0, Ap), where Ap : G X B -4 Band Ap(g, a) = g(a).
We next rephrase Corollary 1.4, and show how it follows from the Expressibility
Theorem.

Corollary 2.10 (The Reconstruction Theorem for Local Movement Systems) For
i = 1,2 let (Bi' Gi ) be a local movement system. Suppose that a : G l ==- G 2 . Then
there is f3;2 a such that f3: M(Bl,Gd ==- M(B 2,G2 ).

=
Proof: Let Pi {var(g)lg E G;}. For a E PI let 9 E G 1 be such that var(g) a, =
and define Il(a) = var(a(g». If hE G 1 and var(h) =
a, then G l 1= <PEq[g,h].
So G 2 F= <pEq[a(g), a(h)]. Hence var(a(h» = var(a(g». This shows that the
definition of 11 does not depend on the choice of g. A similar argument shows that
11 is injective.
If bE P2, then for some 9 E G 2 , b = var(g). Hence b = Il(var(a-l(g»). So
RngCld = P2.
Let a, bE PI and as b. Let g, hE G 1 be such that var(g) = a and var(h) = b.
Hence G 1 F= <p~[g,h]. So G 2 F= <p~[a(g),a(h)], and thus var(a(g» S var(a(h».
This shows that 11 is order preserving.
=
Let a, bE P2 and a S b. Let g, hE G 2 be such that var(g) a and var(h) = b.
Hence G 2 F= <pdg, h]. So G 1 1= <pda- 1(g), a- 1(h)], and thus var(a- 1(g» S
var(a- 1(h». SO-/11(a) S 111 (b). W;have shown that 11 : PI ==- P2 .
By the fact that G l and G2 are locally moving, PI and P2 are dense in BI
and B2 respectively. So 11 extends to an isomorphism I between Bl and B 2 . Let
f3=aU / ·
We show that f3 : M(Bl' G l ) ==- M(B2' G2). It remains to show that f3 preserves
Ap.
Let I E G 1, a E PI and b = I(a). Let g, h E G l be such that var(g) = a
and var(h) = b. So G 1 1= <PAp[j, g, h], and hence G 2 F= <PAp [a(!), a(g), a(h)]. So
a(f)(var(a(g))) = var(a(h». So 1(f)(/(a» = I(b).
Suppose now that a E B1 and 9 E G 1.
132 Matatyahu Rubin

So Ap(g, a) = g(a) = E {g(b) I b E P l n Bt\a}. Hence

t3(Ap(g,a» =t3(E{g(b)lbEPl nBt\a})

= E{t3(g(b)) I b E Pl n Blla}
= E{t3(g)(t3(b)) I b E Pl n Btta }
= E{t3(g)(c) IcE P2 n B21t3(a)}
= t3(g)(t3(a)) = Ap(t3(g), t3(a)). o
Another seemingly interesting corollary of the Expressibility Theorem is the fact
that the class of locally moving groups is elementary. That is, there is a first order
sentence in the language of groups which holds exactly in the locally moving groups.
We state this as Corollary 2.12 below and show how it follows from Theorem 2.5.
The following lemma due to Biichi is the main additional claim needed in the proof
of 2.12.

Lemma 2.11 [3] Let (P, <) be a poset without a minimum. We write a lib if there
is c E P such that c ~ a, b. We write a 1. b to mean a lr b. We say that (P, <) is
separative if for every a, b E P: if a 1: b, then there is c E P such that c :::; a and
c 1. b. Then the following are equivalent.
(1) P is embeddable as a dense subset of a complete BA.
(2) P is separative.

Proof: A complete proof can be found in [7] Chapter 2 Proposition 4.16. Let us
indicate another proof.
The fact that (1) ::::} (2) is trivial.
(2) ::::} (1). Let P be a poset without a minimum which is separative. Let A ~ P
and b E P. We say that b is covered by A, if for every b' ~ b there is a E A such
that b' II a. We say that a subset A ~ P is complete, if for every b E P, if b is
covered by A, then bE A. Let B be the set of complete sets ordered by inclusion.
Claim 1: B is a complete BA.
Claim 2: For a E P let pS.a d~ {b E PI b ~ a}. Then for every a E P, pS.a E B,
and the function at-> pS.a is an embedding of P onto a dense subset of B.
The proofs of both claims are easy. The first proof though, is a bit long. 0

Corollary 2.12 There is a first order sentence 'PLM in the language of groups such
that for every group G: G 1= 'PLM iff G is a locally moving group.

Proof: Note that the fact that the poset P is separative is expressible by a first or-
de!
der sentence. Let G be any group. For 9 E G let V(g) -
{h E GIG 1= 'Ps.[h,g]}. Let V(G) d~ {V(g) I 9 E G - {Ie}}.
Let 'PLM be the conjunction of the following first order statements.
(1) (V(G),~) has no minimal elements.
(2) (V( G),~) is separative.
 Locally Moving Groups and Reconstruction Problems 133

(3) For every 9 E G, if for every hE G, V(h g ) = V(h), then 9 = 1G.

(4) For every h E G - {IG} there is 9 E G - {IG} such that for every kEG - {Ie},
=
if in (V(G), ~), V(k) -L V(h), then V(k g ) V(k),
Note that the conjunction of (1) and (2) says that (V(G),~) is embeddable as
a dense subset of a complete atom less Boolean algebra. (3) says that the action of
G on V(G) by conjugation is faithful. (4) says that the above action is a locally
moving action.
It follows from the Expressibility Theorem that if G is a locally moving group,
then G f= 'PLM·
Clearly, <PLM describes a faithful action of G on a dense subset of some atomless
complete BA B. By 2.8(b), such an action can be extended to an action on B, and
by (4), it is a locally moving action. So if G f= <PLM, then G is locally moving. 0
The rest of this section is devoted to the proof of the Expressibility Theorem. In
what follows, B denotes an atom less complete BA and G denotes a locally moving
subgroup of Aut(B).

Notation 2.13 For a E B let SPG(a) d,;j {g E G I var(g) :::; a}.

n
Lemma 2.14 (a) Let 0 f a < II var(gi); then there is 0 f b < a such that
;=1
n
b· Lg;(b) = O.
;=1
(b) Let a f 0 and n EIN+. Then there is h E SPG(a) such that h n =F!d.
(c) Let 0 =F a:::; var(f)·var(g). Then there is hE SPG(a) such that [fh,gj =F Id.
(d) Let 9 E G and 0 =F a :::; var(g). Then there is k E SPG(a) such that
[k, gj =F Id.
Proof' (a) is proved by a trivial induction on n.
(b) We prove by induction on n that there are h E Sp G (a) and 0 f b :::; a such
that b, h(b), ... , hn(b) are pairwise disjoint. For n = 1, let h E SPG(a) - {Id}.
So there is 0 =F b :::; var( h) such that b . h( b) = O. Suppose the claim is true for
n. Let h E Sp aC a) and 0 f b :::; a be as assured by the induction hypothesis. If
h n+1 [b =F !d, let 0 =F c :::; b be such that c· hn+l( c) = O. So hand c are as required.
Suppose that hn+l[b = Id. Let k E SPG(b) - {Id} and 0 =F c :::; b be such that
c· k(c) = O. Let 9 = kh. So for i::::; n, gi(c) = hi(c):::; hi(b). Hence C, . . . ,gn(c) are
pairwise disjoint. Since gn+l(c) = (kh)n+l(c) = kh(kh)n(c) = kh(hn(c» = k(c),
then gn+l(c) . c = O. Since gn+l(c)::::; b, for every 1::::; i::::; n, gn+l(c)· iCc) = o.
(c) Let 0 =F a:::; var(f)· var(g). If [j,g] f!d, then h =!d is as required. So
we assume that [j,gj = !d. Let 0 "I- bl ::::; a be such that bl · (f(bd + g(bJ) = O.
If bl . var(fg) "I- 0, then let 0 f b ::::; bl be such that b . fg(b) = o. Otherwise, let
b = bl . Let h E SPG(b) be such that h 2 f Id. Let c ::::; b be such that c, h(c), h 2 (c)
are distinct.
134 Matatyahu Rubin

Then,

(1) gfh(h(e)) = ghfh- 1h(e) = ghf(e) = gf(e).

(2) fhg(h(e» = hfh- 1gh(c) = hfgh(e).

Case 1: fg)b -1 Id. Then by the choice of b, fg(b)·b = O. Hence in (2) we obtain
that hfgh(e) = fg(h(e)) gf(h(c». Since gf(h(e» -1 gf(e),
fh and 9 do not commute.
Case 2: fglb = Id. Then in (1) gf(e) = e and in (2) hfgh(e) = h 2 (e). So 9 and
fh do not commute.
(d) Let 0 -1 a :::; var(g). Let f E S'PG(a) - {Id}. By part (c), there is h E
S'PG(var(J» such that [r,gj-1 Id. It follows that fh is as required. 0

Notation 2.15 Let Cc(!) = {g E G I [g, fJ = Id}. We abbreviate CG(J) by C(J).

The next lemma says that for f E Aut( B) and a E B, no sum of a pairwise
disjoint collectioll of TI + 1 images fk i (a) can be covered by any sum of TI images
hi(a), when the hi'S commute with f·

Lemma 2.16 Let ko, .. ·, k n Ell, f E Aut(B) and a E B. Suppose that rO(a), ... ,
fk n (a) are pairwise disjoint. Then for every hi, ... , h n E C(J) and 0 -1 b :::; a,

L i L"
11

fk, (b) hi(a).

;=0 ;=1

Proof: By induction on TI. Let n = 1. If fko(b) + fkl(b) :::; h 1 (a), then 0 -1 b':::;
f-k o(h 1(a»· f-k.(h 1(a». Hence since hI E C(J), 0 -1 f-ko(a)· f-k 1 (a). Apply
fk o+kl to both sides. So r'(a)· fko(a) -10, a contradiction.
Suppose the claim is true for TI. Let k o, ... , k n +1 , hI,"" hn+1, a and b be as in
n+1
the lemma. If h n +1 (a)· Lfk'(b) = 0, then by the induction hypothesis
;=0

n+1 n+1
L h;(a) t L fk'(b).
;=1 ;=0

Suppose otherwise, and let j be such that h,,+1 (a) . fk, (b) d;J c f. O. We may
assume that J = n + 1. We will define by induction for 0 .:::; i .:::; TI, 0 f. b; :::; b such
that (i) h n +1(a)·r'(b;) = 0, and such that (ii) bi +1 :::; b;. Let b' = f-k n +1 (e). Since
t
o f. b' :::; a, and by the case TI = 1, h"+l(a) fko(b') + fk n +1 (b') = fko(b') + e.
Since h"+I(a) 2: e, h"+I(a) t
fko(b').
Let bo = f-ko(Jko(b') - hn+1(a». It follows that 0 -1 bo .:::; b. Also, (i) holds.
Suppose bi has been defined. So bi .:::; b'. Hence hn+l(a) t fki+1(b;) + fk n +1 (b;).
Locally Moving Groups and Reconstruction Problems 135

But hn+1(a) 2: flc n +1 (b i ). Let bi+1 = f-Ic;+l(Jlci+l(bi) - hn+l(a». So (i) and (ii)
hold. n
We obtain that hn +1 (a)· I)Ic, (b n ) = O. By the external induction hypothesis,
i=O

n n
Lhi(a) l Lflc;(bn )
i=1 i=O

and so
n+1 n
L h;(a) l L flci(b n ).
;=1 ;=0

Hence
n+l n+l
Lhi(a) l Lflc;(b). o
i=1 i=O

='v'g«[g, f] f- Jd)
Lemma 2.17 Let
'Pl(J, 1') ---+

(3h,h E C(J'»«[g, h, 12] f- Jd) 1\ ([[g, h, 12], f'] = Jd))).

(aJ For every local movement system (B, G) and f, f' E G: if
var(J) . var(J') = 0, then G F 'Pdf,J']·
(bJ For every local movement system (B, G) and f, f' E G: if
var(J) . var«(J')12) f- 0, then G ~ 'Pdf, f'].

Proof' (a) Let (B, G) be a local movement system. Let f, f' E G and suppose
that var(J) . var(J') = O. Let 9 E G be such that [g,J] f- Id.
So var(g) . var(J) f- O. By proposition 2.3(b), there is
a E B[(var(J) - {OJ) such that g(a) . a = O. Let h E SPG(a) - {Id}. Hence
h E C(J'). Let gl = [g, ft]· So gl = ghg- 1f11 = (h)9 . f11. Recall that
g(var(h» . var(Jt) = O. So var(9d =
var(J!) + g(var(h» and gl(var(h» =
var(h)·
Let b E B[(var(J!) - {OJ) be such that gl(b)· b = O. Let 12 E Spdb) - {Id}.
Since b· var(J') = 0, 12 E C(J'). Let 92 = [91,12]. So g2 = f~' . f:;l. g1( var(h» .
var(h) = O. So var(g2) = var(h) + g1 (var(h» f- O. Also var(h) ::; var(Jd and
gl (var(JI) = var(h). So var(g2) ::; var(h). Hence var(g2) . vaT(J') = O. So
92 E C(J'). We have shown that G F 'PI [f,J']·
(b) Suppose that a d~ var(J) . var«(J/)12) f- O. So for every i = 1, ... ,4,
a::; var«(J')i). Hence by Lemma 2.14(a), there is b E B[a - {OJ such that
4
b . L(J')i(b) = O.
;=1
136 Matatyahu Rubin

Hence {(f')i(b) I 0 :::; i :::; 4} are pairwise disjoint. By Lemma 2.14(d), there is
g E SPG(b) such that [g, fl "# !d.
f'l
de]
Let h, h E C(f') be such that g2 = [[g, hl, hl "# Id. We show that [g2, "#
Id. Now,
g2 = [g. (y-1)h, hl = g(g-l)h . ((g. (g-l)h )-1)12
= g(g-l)h . (gil g-l)h = g(y-1)h ghh (g-l)12.
So 0"# var(g2) :::; b + h(b) + hh(b) + h(b).
Let hE {Id,h,hh,h} be such that c d;j var(g2)' h(b)"# O. Since {(f')i(b) I
0:::; i :::; 4} is pariwise disjoint, and hE C(f'),
{(f')i(h(b)) 10:::; i :::; 4} is pairwise disjoint.
Suppose by contradiction that [g2, I'l = !d. Hence for every i = 0, ... ,4,
(g2)(jI)' = g2. Recall also that var(g2) 2: c. So
4 4 4
(1) var(g2) = L var((gd]I)') = 2::::(f')i( var(g2)) 2: 2:::: (f')i (c).
i=O i=O i=O
But var(g2) :::; b + h(b) + h(b) + hh(b). By lemma 2.16, and since h- 1, hh- 1,
hh- 1 and hhh- 1 commute with I',
4
I)J')i(c) 1. h- 1 (h(b)) + hh-l(h(b)) + hh-l(h(b)) + hIth-l(h(b)) =
;=0

b + It(b) + h(b) + hIt(b) 2: var(g2).

1

Hence 2::::(f')i(c) 1. var(g2). This contradicts (1), and so [g2, J'l "# Id. We have
;=0
shown that G ft: !pdf, I'l· o

Definition 2.18 Let D1 (f) d;j {I' I G F !pdf, I'D and

V(f) d;j C({(f')121 I' E DI(f)}).

Proposition 2.19 For every f E G, V(f) = SPG(var(f)).

Proof: Let g E SPG(var(f)). Let f' E D1(f). So var((f')12). var(f) = 0,
and hence var(g) . var((f')12) = O. Hence [g, (f')12] = !d. We have shown that
Sp cC var(f)) ~ V(f).
Next let g E G - Sp d var(f)). Hence a = var(g) - var(f) "# O. By Lemma
de]

2.14(b), there is f' E SPG(a) such that (f')12 "# Id. By Lemma 2.14(c), there is
h E SPG( var((f')12)) such that [((f')12)h, g] "# Id. Now var(f') . var(f) = 0 and
var(h) :::; var((f')12) :::; var(f'), so var((f')h) = var(f'), and hence var((f')h) .
var(f) = O. It follows that (f')h E D1(f). So g 1:. V(f). We have shown that
V(f) ~ SPG(var(f)). 0
Locally Moving Groups and Reconstruction Problems 137

Proof of Theorem 2.5:

Let 'P<(f,g) == V(f) ~ V(g). Clearly, 'P<(f,g) can be written as a first order
formula. -If var(f) ::; var(g), then by the ~bove proposition, G F= 'Pdf,g]. If
on the other hand, var(f) 10 var(g), then since G is locally moving, there is h E
SPeC var(f)) - SPeC var(g)). By the above proposition, V(f) C£ V(g). So 'P<!o(f, g)
is as required.
Let 'PEq(f, g) == 'P<!o(f, g) 1\ 'P<!o(g,f) and 'PAp(f, g, h) == 'PEq(gJ, h). It is trivial
that 'PEq and 'PAp are as required. 0

3 Reconstruction of topological spaces

In this section we describe two applications of the Reconstruction Theorem (Corol-

lary 2.10) in topology. The first one, Theorem 3.1 appeared in [12]. The second
one, Theorem 3.8 is new.
We shall use the the notions defined in Definition 1.7. We shall also use Propo-
sition 1.8. Its proof is very easy, and is left to the reader. By Proposition 1.8(a),
H(X) can be regarded as a subgroup of Aut(Ro(X)).
Let K be a class of pairs of the form (X, G), where X is a topological space and
G is a subgroup of the group H(X) of all self-homeomorphisms of X. We say that
K is faithful if for every (Xl, Gl) E K, 1= 1,2 and every isomorphism 'P between
G 1 and G 2 there is a homeomorphism T : Xl ==' X 2 which induces <p, that is, for
every 9 E G l , 'P(g) = Tog 0 T- 1 .
For G ~ H(X) and x E X, let G(x) d~ {g(x) I 9 E G}. A set D ~ X is
somewhere dense if there is a nonempty open set U such that D n U is dense in U.
For an open set V ~ X, let Sp e(V) d~ {g E G I var(gRo) ~ V}. We leave it to the
reader to check that var(gRo) ~ V iff gl(X - V) = [d.
Let K Lce d~ {( X, G) I X is a locally compact Hausdorff space, and for
every open V ~ X and x E V, SpcCV)(x) is somewhere dense}.
The following theorem appears in [12]. It is the conjunction of parts (a), (b)
and ( c) of theorem 3.5 there.

Theorem 3.1 K Lce is faithful.

In the proof of the second result in this section we shall need a theorem somewhat
stronger than 3.1. We next define the notions needed in the statement of this
theorem.

Definition 3.2 Let X be a topological space.

(a) For x E X let Nbrx(x) denote the set of regular open neighborhoods ofx in
X. The superscript X is usually omitted.
(b) Let G::; H(X). The set of densely conjugated points of (X, G) is defined as
Dc(X,G) = {x E X I G(x) is somewhere dense}.
138 Matatyahu Rubin

The set of good points of (X, G) is defined as

Gd(X,G) = {x E X I for every U E Nbr(x), Spa(U)(x) is somewhere dense}.
(c) X is regionally compact if U{T ~ X I T is open and cl(T) is compact} is
dense in X.
(d) Let
J{RCM d;J {(X, G) I X is Hausdorff and regionally compact, G ::; H(X) is a
locally moving subgroup of Aut(Ro(X» and Dc(X, G) = Gd(X, Gn.
(e) For (X, G) E J{RCM let
Mdc(X, G) d;J (Ro(X), Dc(X, G), G;::;, f, 0, Ap), where
f = {(x, U) I x E Dc(X), U E Ro(X) and x E U}.

Note that by definition, Gd(X, G) ~ Dc(X, G). Note also the following trivial
observations.

Proposition 3.3 (a) Let X be Hausdorff and regionally compact. Then for every
non empty open U ~ X, there is nonempty V E Ro (X) such thai cl(V) ~ U.
(b) J{LCG = {(X,G) E J{RCM I Gd(X,G) = X}.

Proof: The proof of part (a) is similar to the proof that a locally compact
Hausdorff space is regular. The proof of part (b) is easy. 0

We can now state the strengthening of Theorem 3.1.

Theorem 3.4 For I = 1,2 let (XI, GI) E J{RC M, and 'P be an isomorphism between
G l and G2. Then there is 'IjJ: Mdc(Xl,Gd =:: Mdc(X 2 ,G 2 ) such that 'IjJ extends 'P.

Let us first see how Theorem 3.1 follows from Theorem 3.4.
Proof of 3.1: We assume Theorem 3.4. For 1= 1,2 let (XI, GI) E J{LCG and 'P
be an isomorphism between G l and G 2 . By Theorem 3.4, there is

Then DC(XI' Gt) = XI. So 'IjJ!X l is a bijection between Xl and X 2 . Since 'IjJ
preserves E, for every U E Ro(Xd, {'IjJ(x) I x E U} = 'IjJ(U). So 'IjJ!X l takes the basis
Ro(X l ) of Xl to the basis Ro(X 2 ) of X 2. So 'IjJ!X l : Xl =:: X 2 . Since 'IjJ preserves Ap
and 'IjJ extends 'P, for every 9 E G l and U E Ro(X 2 ), 'P(g)(U) = g(t/ifRo(X,))(U), and
since for every x E X 2 , {x} = nNbr(x), 'P(g)(x) = g(t/ifx,)(x). So 'IjJ!X l induces 'P.
o
The following notions will be needed in the proof of Theorem 3.4.

Definition 3.5 (a) A subset p of a BA B is called an ultrafilter if: (i) a ct. p; (ii) if
al, ... , an E p, then rr~=l ai E p; (iii) if a E p and b ~ a, then b E p; and (iv) pis
not properly contained in a subset of B satisfying (i)-(iii).
The set of ultrafilters of B is denoted by Uli(B).
Locally Moving Groups and Reconstruction Problems 139

A subset q of B has the finite intersection property if for every finite (J' ~ q,
n t- O. Note that by Zorn's lemma, every subset of B with the finite intersection
(J'

property is contained in an ultrafilter.

Let (X,G) E KRCM·
(b) We say that an ultrafilter p in Ro(X) is densely conjugated if there is W E
Ro(X) - {0} such that for every V E Ro(X)JW - {0} there is 9 E G such that
g(V) E p. Let Dcu(X, G) denote the set of densely conjugated ultrafilters.
(c) For p E Ult(Ro(X)), let Ap d~ n{cl(V) I V E pl. If Ap is a singleton, let xp
be such that Ap = {xp}.
Theorem 3.4 follows from a certain "Expressibility Theorem" in very much the
same way that the Reconstruction Theorem for local movement systems followed
from the Expressibility Theorem for locally moving groups. That is, we shall find
certain objects in M(Ro(X), G) which will represent the densely conjugated points
of X. In fact, each densely conjugated point x E X will be represented by any
densely conjugated ultrafilter p such that xp = x. We shall then show that the
fact that two densely conjugated ultrafilters p and q represent the same point x, is
equivalent to a property of (p, q) expressible in M(Ro(X), G). Similarly the fact
that xp E U will be shown to be equivalent to a property of (p, U) expressible in
M(Ro(X), G).
The above explanation may be a little vague, because ultrafilters are not mem-
bers of M(Ro(X), G) but rather subsets of M(Ro(X), G). We shall later formalize
the situation by referring to formulas in second order logic.

Proposition 3.6 Let (X,G) E KRCM and p E Ult(Ro(X).

(a) lAp I :s; 1.
(b) If Ap is a singleton, then p 2 Nbr(x p).
(c) If x E X and p 2 Nbr(x), then Ap is a singlton, and xp = x.
(d) Ifp is densely conjugated, then Ap is a singleton, and xp E Dc(X,G).
(e) If x E Dc(X, G) and p 2 Nbr(x), then p E Dcu (X, G).

Proof" Part (a) follows from the fact that X is a Hausdorff space.
(b) Suppose that Ap = {x}. Then pU Nbr(x) has the finite intersection property.
By the maximality of p, p 2 Nbr(x p).
(c) Let U E p. So X - cl(U) (j. p. So X - cl(U) (j. Nbr(x). So x E cl(U).
(d) Let p be densely conjugated. Let W E Ro (X) - {0} be such that for every
V E Ro(X)JW - {0} there is 9 E G such that g(V) E p. Let V E Ro(X)JW - {0}
be such that cl(V) is compact, and let 9 E G be such that V' d~ g(V) E p. So
cl(V') is compact. Hence Ap = n{cl(U) n cl(V') I U E p} t- 0. So by part (a), Ap
is a singlton.
We show that G(xp) n W is dense in W. Let U E Ro(X)IW - {0}. Let
V E Ro(X) - {0} be such that cl(V) ~ U. For some 9 E G, g(V) E p. So
xp E cl(g(V». Hence g-l(xp) E cl(V) ~ U. So G(xp) n W is dense in W.
140 Matatyahu Rubin

(e) Let x E Dc(X, G) and p 2 Nbr(x). Let W E Ro(X) - {0} be such that
G(x) n W is dense in W. Let V E Ro(X)[W - {0}. Then there is 9 E G such that
g(x) E V. So g-l(V) E Nbr(x) ~ p. Hence p is densely conjugated. 0

We shall next present three formulas in second order logic with variables ranging
over subsets of M(Ro(X), G). Note first that the fact that p is an ultrafilter can be
expressed by such a formula. That is, p is an ultrafilter iff it satisfies in M(Ro(X), G)
the formula 'lj;UIt (p) defined as follows:
de!
'lj;Ult(P) == (p ~ Ro(X)) /\ (0 rf. p)/\
(VU E Ro(X))(VV E Ro(X))((U, V E p) --> (U· V E p))/\
(VU E Ro(X))(VV E Ro(X))((U E P /\ U :<:; V) --+ (V E p))/\
(VU E Ro(X))((U E p) V (-U E p));
de!
'lj;Dc(P) == 'lj;Ult(p)/\
(3W E Ro(X) - {0} )(VV E Ro(X)[W - {0} )(3g E G)
(g(V) E p);
de!
'lj;Eq(p,q) == -,(3U,Ua Ep-q)(VVERo(X)[Ua-{0})(3gESpc(U))
(V E f(p));
de!
and 'lj;e (p, U) == (Vq)(( 'lj;Dc( q) /\ 'lj;Eq(p, q)) --> (U E q)).
The important property of these formulas is that they involve only the relations
and functions of the structure M(Ro(X), G). So if 1/J is any of the above formulas,
T: M(Ro(Xd,GI) == M(Ro(X 2 ),G 2 )) and p is a subset of M(Ro(Xd,Gd, then
M(Ro(Xd, Gd F= 1/J[P] iff M(Ro(X 2 ), G 2 ) F= 1j)[T(p)].

Proposition 3.7 Let (X, G) E f{RCM, p, q E Ult(Ro(X)) and U E Ro(X). Then

(a) M(Ro(X), G) F= 1/JDc[P] iff P E Dcu(X, G).
(b) Ifp,q E Dcu(X,G), then xp and Xq exist,
and M(Ro(X), G) F= 1/JEq[P, q) iff xp = x q .
(c) Ifp E DCll(X,G), then M(Ro(X),G) F= 1/Je[P,U] iffxp E U.

Proof: Part (a) is completely trivial because 1/' Dc (p) just stat.es the definition of
being a densely conjugated ultrafilter.
(b) Let p, q E Dc (X, G). It is proved in Proposition 3 .6( d) that xp and x q exist.
Suppose first that Xp = Xq . Suppose by contradiction that M(Ro(X), G) ~
1/JEq[P, q]. Let U and Ua be as assured by -,1/JEq[P, q). So U, Ua rf. q, and hence
U, Ua rf. Nbr(x p). Let V E Ro(X) - {0} be such that cl(V) ~ Ua. Let 9 E Spc(U).
Since xp rf. u, g(x p) = xp rf. Uo, and hence g(x p) rf. cl(V). But xg(p) = g(xp), so
V rf. g(p). This means that U and Uo do not satisfy the requirement of -'rPEq[P, q],
a contradiction.
Suppose next that xp i- xq. By Proposition 3.6(d), xp E Dc(X, G). Since
(X, G) E f{RcM, xp E Gd(X, G). Let U E Nbr(x p) be such that Xq rf. cl(U). Let
Uo ~ U be such that Spa(U)(xp)nUa is dense dense in Ua. Let V E Ro(X)[Ua-{0}.
Locally Moving Groups and Reconstruction Problems 141

Then there is g E SpG(U) such that g(xp) E V. So V E Nbr(g(xp)) = Nbr(xg(p)) ~

g(p). Hence U and Uo are as required in -'1f!Eq[P, q].
(c) Let p E Dcu(X,G) and U E Ro(X). Suppose first that xp E U. Let q be
such that M(Ro(X), G) P= 1f!Dc[q] II 1f!Eq[P, q]. By parts (a) and (b), q E Dcu(X, G)
and xp = xq. So U E Nbr(x q) ~ q. We have shown that M(Ro(X), G) P= 1f!£[p, U].
Suppose next that xp ~ U. So Nbr(xp) U {-U} has the finite intersection
property. Let q be an ultrafilter containing Nbr(xp)U{-U}. By Proposition 3.6(c),
Xq = xp. Also, xp E Dc(X, G), hence by by Proposition 3.6(d), q E Dcu(X, G), and
so M(Ro(X), G) P= 1f!Dc[q]. By part (b), M(Ro(X), G) P= 1f!Eq[P, q]. But U ~ q. So
M(Ro(X), G) pt: 1f!£[p, U]. 0

Proof of Theorem 3.4: For I = 1,2 let (Xl, Gl) E /{RCM, and 'P be an iso-
morphism between G t and G2. Since (RO(Xl), Gl) are local movement systems,
by 2.10, there is a ;2 'P such that a: M(Ro(XI), Gd :::! M(Ro(X 2 ), G2). We de-
fine (3 : DC(Xl' Gd --+ DC(X2' G2). Let x E DC(Xl' Gt). Let p be such that
Nbr(x) ~ p E Ult(Ro(Xt)). Then by 3.6(d), 3.7(a) and 3.6(c), Xa(p) exists and it
belongs to DC(X2' G 2 ). Let (3(x) = Xa(p)·
The proof that {3 is well-defined and that a U (3 is an isomorphism
between Mdc(X t , Cd and Mdc(X 2 , C2) is similar to the proof of the parallel parts
in the Reconstruction Theorem for Local Movement Systems (Corollary 2.10). 0

Foliated Spaces
Previous versions of the theorem on locally moving groups were not general
enough to handle the homeomorphism groups associated with certain foliated spaces.
The simplest such example was the torus with an irrational foliation.
Let us describe the setting. We represent the torus T as [0,1) X [0, 1). Let frc(a)
denote the fractional part of the real number a. Let cvr(a, b) d~ (Jrc(a),Jrc(b)). So
cvr is the covering map from IR 2 to T. A set C of straight lines in IR 2 is called a
parallel family if every two distinct members of C are parallel. For a parallel family
C let F(C) d~ {cvr(L) I L E £} and T c d~ U F(C). Note that if C consists of a
single line with an irrational slope, then T c is dense in T. We consider two possible
groups. Let
dej
Ho(Tt:.) = {h E H(Te.) I for every L E F(C), h(L) = L}; and
HI(Tt:.) d~ {h E H(Te.) I for every L E F(C), h(L) E F(C)}.
Note that if the members of C have an irrational slope, and T c # T, then HI ( T e.) =
H( Te.).
As a corollary of a more general statement, we shall obtain the following result.

Theorem 3.8 For i = 1,2 let C i be parallel families and mi E {O, I}. Suppose that
'P: Hm,(Tc,):::! H m2 (Tc 2)· Then there is T: T c , :::! Tc. such that T induces 'P,
and either ml = m2 or Hml (T c,) = H m2 ( T c,).
142 Matatyahu Rubin

We shall next define a notion of a foliated space. The reconstruction problem

for this class of spaces is still open. We can only prove a special case. Another
important special case was proved by W. Ling in [8]. He proved the reconstruction
theorem for Euclidean manifolds with a foliation. His method is entirely different
from the method presented here. Ling's proof requires that the space in question
be locally compact and connected.

Definition 3.9 (a) Let E and F denote normed vector spaces over IR. BE, BE
and SE denote respectively the unit open ball, the unit closed ball and the unit
sphere of E. BE(x, r) denotes the open ball with center at x and radius r. We use
the same notation for B. Let BE,F and BE,F denote respectively BE x BE and
BE x BE. Let ",(E,F)d;d {(y, Zl), (y, Z2)) lyE BE, Zl, Z2 E BF}.
(b) Let o:(y, z) = (O:l(y), 0:2(y, z)) be a map with the following properties:
(1) Dom(o:), Rng(o:) ~ BE,F; (2) 0: : Dom(o:) ~ Rng(o:); and
(:~) cq : Dom(o:d ~ Rng(o:J). Then 0: is called an (E, F)-selfmap.
(c) Let X be a topological space, and, : BE,F --+ X be a map with the following
properties: (1),: BE,F ~ Rng(f); (2) ,(BE,F) is closed in X; and (3) ,(BE,F) is
open in X. Then, is called a normal (E, F)-map.
(d) Let M = (X, r) be as follows: (1) X is a topological space; (2) f is a set of
normal (E, F)-maps, and for every, E f, Rng(f) ~ X; (3) For every x E X there
is, E f such that x E ,(BE,F); (4) For every (3" E f, (3-1 0 , is an (E, F)-selfmap;
and (5) There is fo ~ f such that Ifol S No, and for every ~( E f there is ,0 E fo
such that Rng(f) ~ Rng(fo). Then M is called a foliated mamfold (PI-manifold)
based on (E, F).
We call fa a dominating set for r, and in fact, the Fl-manifold obtained from
r ° is identical to the one obtained from r.
(e) Let M = (X,1') be an FI-manifold based on (E, F). For, E r
let "', = {(,( u), ,( v)) I U ",(E,F) v}. Let ",M be the transitive closure of
U{ "', I, E r}.
(f) Let M = (X o, r) be an Fl-manifold, and F = (X, "') be such that: (1)
X ~ Xo; (2) for every x E X and Y ",M X, Y E X; (3) '" = ",M [X; and (4) for
every, E f, ,-1(X) is dense in BE,F.
Then F is said to be an FI-subspace of M. We say tlmt F is an FI-space if it is an
FI-subspace of some FI-manifold M.
(g) For a topological space Z and U ~ Z, let bd z ([1) denote the boundary of
U in Z. We say that a topological space Z is thin if for every z E Z there are
U E Nbr(z) and V E Ro(Z)[U such that bdu(V) = {z}.
(h) Let Pt:,F be the projection of BE,F on jjE. Let F and M be as in (f). For
, E r let Y'; = PE,F(,-l(X»). We say that F is thin, if for every, E r, Y'; is
thin.
(i) For a foliated space F = (X. "'), let

H1(F) d~ {h E H(X) I h preserves "'},

 Locally Moving Croups and Reconstruction Problems 143

and
de]
Ho(F) = {h E H(F) I (VX E X)(h(x) '" X)}.
Theorem 3.10 For 1= 1,2 let F/ be thin FI-spaces based on E/, F/, and
cp : Hml (F1 ) ~ H m2 (F2). Then there is T : Fl ~ F2 such that T induces cp.

We shall prove here the special case when the F/'s are finite dimensional. In
order to prove the full theorem, it is necessary to first prove the reconstruction
theorem for infinite dimensional normed spaces. This reconstruction theorem is
proved in [12]. See Part II of Section 3 there, and in particular Corollaries 3.20 and
3.2l.
From here on F = (X, "') denotes a foliated space, and H is either H1(F) or
Ho(F). Let M = (Xo, r) denote an Fl-manifold such that F is an Fl-subspace of
M. We assume that M is based on E, F.
Proposition 3.11 (Ro (X), H) is a local movement system.

Proof: The easy proof is left to the reader. o

We shall represent the points of X by some objects of M(Ro(X), H). This is
quite similar to the way that good ultrafilters represented points when we dealt with
KLCG. In the present case however, the representing objects are more complex, and
the representation requires more preparation.

Proposition 3.12 (a) Let 1,6 E r. Then

!(Rng(l) n Rng(6» = "'6 !(Rng(l) n Rng(6».
"'-y
(b) Let 1 E rand L E Xo/ ",M. Then L n Rng(l) is the unIOn of < No
equivalence classes of "''"(.

Proof: (a) follows trivially from the definition of an Fl-manifold.

(b) follows easily from Part (a) and the fact that r is dominated by a countable
set. 0

Remark: We do not know how to get rid of the artificial assumption that r
has a countable dominating set.
It will be convenient to assume that r is closed in the following sense. Suppose
that f3E (Yo, r) x f3F (zo, s) ~ f3E,F. Let a : f3 E ,F ---+ f3E (y, r) x f3F (z, s) be defined
by a«y, z» = (YO + ~y, Zo + ~z). We assume that if 1 E f, then loa E f.
Definition 3.13 Let U E Ro(X).
(a) Let INV(U) = {V E Ro(X)[U I (Vg E Sp H(U»(g(V) = V)}.
(b) For V E INV(U) and 9 E SPH(U) let L(V;U) = bdU(V),
gV,u = grL(V; U), C(V; U) = {gV,U I 9 E Sp H(U)} and
Gld(V; U) = {g E Sp H(U) I gv,u = Id}.
(c) Let H(E, F) = {a E H(f3 E ,F) I a is an (E, F)-selfmap and
a[(f3E,F _ BE,F) = Id}.
144 Matatyahu Rubin

(d) For I E r let B~ = Rngb) n X and B~ = I(BE,F) n X. We shall usually

omit the superscript :F.

Proposition 3.14 (a) For IE rand h E H(E, F) let

h, = 10 hOI-lIB, U IdI(X - B,) and H, = {h, I h E H(E, F)}. Then H, ~
Sp H(B,),
(b) Let I E r. Let L' ~ Band L = L' x B . Let h E H be such
-E de! -F

that hb(L)) = I(L) and hli(V x SF) = Id. Then for every x E I(L), hex) "", x.
In particular, for every hE SPH(B,) and x E B" hex) "", x.
(c) Let U E Ro(X) and V E INV(U) - {0}. Then (L(V; U), G(V; U)) is a local
movement system.

Proof: (a) It is trivial that H, ~ Sp H(B,).

(b) For notational simplicity we assume that I = I d. Let L' and h be as in (b).
Suppose by contradiction that x E L is such that hex) 7-, x. Let x = (y, z) and
=
h(x) (y', z'). Hence y f:. y'. Let Zl E SF and Xl = =
(y, zt). Then h(xd Xl. Let
J be the line segment connecting x and Xl. A d;j PE,F (h(J)) :3 y, y'. SO IAI = 2 ND .
SO h(J) intersects 2ND equivalence classes of "",' For every u E h(J), u "" Xl. So
xd"" intersects 2ND equivalence classes of "",' This contradicts Proposition 3.12.
(c) Let L = L(V; U) and G = G(V; U). Since V E INV(U), L is invariant under
SPH(U), So G is a group. Let X ELand T E NbrL(x). We show that there is
g E Sp dT) such that g( x) f:. x. Let I E r be such that x E B,. For notational
simplicity we assume that I = Id. We may also assume that B, ~ U, and that
B, n L ~ T. Since L is invariant under Sp H(U), L n B, has the form A x BF,
where A ~ BE. Obviously, there is hE H(E,F) such that hex) f:. x. It follows
de!
that g == h,IL E G, and that gl(L - T) = Id. So g is as required. 0

Proposition 3.15 (a) If x "" y, then there is an arc J ~ X connecting x and y

such that for every z E J, x"" z.
(b) Let h E Hand W E Ro(X) be such that g d;j h!W U Id!(X - W) E H(X).
Then g E H.
Let U E Ro(X) - {0} and V E INV(U).
(c) The function g/GId(V; U) f-+ gIL(V; U) is an isomorphism between
Sp H(U)/GId(V; U) and G(V; U).
(d) GId(V; U) = {gh I g E Sp H(V), hE Sp H(U _Ro V)}.
(_Ro denotes the difference operation in Ro(X)).

ReIllark: Part (d) has a real shortcoming. It does not transfer to the case when
H is a group of diffeomorphisms. We did not find a way to fix this at this time.
However, the method seems to work for the case that H is a group of bi-Lipschitz
homeomorphisms.
Proof: The easy proofs of parts (a) and (c) are left to the reader.
Locally Moving Groups and Reconstruction Problems 145

(b) Let h, Wand 9 be as in part (b). We show that 9 E H. If H = Ho(F), then

clearly for every x E X, g(x) ~ x. So 9 E H.
Suppose that H = HI (F). Suppose by contradiction that 9 r{. H. Then by
replacing 9 by g-1 if needed, we may assume that there are x, y E X such that
x ~ y and g(x) f g(y). We may assume that x E Wand y EX - W. So x f hex).
Let J be an arc connecting x and y such that for every z E J, z ~ x. Then
h(J) n J = 0. Also g(J) ~ J u h(J), g(x) E g(J) n h(J) and y E g(J) n J. So g(J)
is not connected, a contradiction.
(d) Let f E GId(U; V). We define 9 = (ltv) U (Idl(X - V)) and h =
fg-l.
Clearly 9 E H(X). So by part (b), 9 E H. Hence h E H. Obviously, 9 E SPH(V)
and h E Sp H(U _Ro V). So GId(V; U) ~ {gh I 9 E Sp H(V), hE Sp H(U _Ro V)}.
The inclusion in the other direction is trivial. 0

By combining parts (c) and (d) of Proposition 3.15 together with part (b) of
Proposition 3.14, we obtain an expressibility result. Loosely speaking, it says that
there is a first order formula in the language of M(Ro(X), H) that says about U E
M(Ro(X),H), V E INV(U)- {0} and g,h E SPH(U) that var(gv,u)::; var(hv,u).
Similarly, there are formulas expressing the facts that var(gv,u) = var( hv,u) and
that fv,u( var(gv,u» = var( hv,u). As expected, these expressibilty facts translate
into certain intermediate reconstruction results.
There is a formal framework in which these expressibility and reconstruction
facts can be stated neatly and precisely. However, we decided to spare the reader
from these formalities. A detailed exposition of this framework is presented e.g.
in [15) Chapter 1. We shall proceed somewhat informally, relying on the reader's
intuition to translate expressibility facts into reconstruction results.

Definition 3.16 Let U E M(Ro(X), H), V E INV(U) - {0} and L = L(V; U).
We say that (U, V) is a usable pair if for every T E Ro(L) - {0} there are "I E r,
y E BE, Z E BF and r > 0 such that
Tn 'Y(BE(y, r) x BF(z, r» = 'Y( {y} x BF(z, r».
We shall prove three main facts about usable pairs. (1) The property of being a
usable pair is expressible in M(Ro(X), H). (2) If F is finite dimensional, and (U, V)
is a usable pair, then (L(V; U), C(V; U») E /{RCM, and Dc(L(V; U), C(V; U» is
dense in L(V; U). (3) For every x E X there is a usable pair (U, V) such that
x E Dc(L(V; U».

Definition 3.17 (a) Let B be a complete atomless BA and C ::; Aut(B). Let
a, bE B - {OJ. Then a --<!,G b, if for every c E B[b - {OJ there is 9 E SPG(b) such
that g( a) ::; c.
(b) let 1/Jsml be the following sentence in the language of M(B,C).
1/J.,ml == (Va E B - {OJ )(3b, c E B[a - {OJ )(b --<!, c).

Lemma 3.18 (aJ Let U E Ro(X) and V E INV(U) - {0}. Then the following
conditions are equivalent.
146 Matatyahu Rubin

(1) (U, V) is usable;

(2) M(Ro(L(V; U)), C(V; U)) F= ~sml.
(b) There is a first order formula ~;ml(U, V) in the language of M(Ro(X), H)
such that for every U E Ro(X) and V E fNV(U) - {0},
M(Ro(X), H) F= ~;ml[U, V] iff M(Ro(L(V; U)), C(V; U)) F= ~sml.
Proof: Part (b) follows trivially from 3.15(c) and (d) and 3.14(b).
(a) Let U and V be as in part (a). Let L = L(Vi U) and C = C(V; U). It
follows easily from 3.14(a) that if (U, V) is usable, then M(Ro(L), C) F= ~sml.
Suppose next that M(Ro(L),C) F= ~sml. Let T E Ro(L) - {0}. Let T* E
Ro(X)[U be such that T = T* n L. Let 'Y E r be such that B"'I n T I- 0. For
notational simplicity we assume that 'Y = fd. We may assume that B"'I <;;;; T*.
Hence since V E fNV(U), there are V' <;;;; BE and L' <;;;; BE such that V n B"'I =
V' X BF and Tn B"'I = L n B"'I = L' X BF. Since M(Ro(L), C) F= ~.ml there are
R, S E Ro(L)I(T n B"'I) - {0} such that R -<.G S.
Suppose by contradiction that IPE,F(S)I > 1. Let SI, S2 E Ro(L)!S - {0}
be such that PE,F(SI) n PE,F(S2) = 0. Since R -<. 5, there are gi E SpaCS),
i = 1,2, such that gi(R) <;;;; Si. Clearly, gi E SPG(LnB"'I)' By the definition ofC,
there are hi E H such that gi = h;fL. So hi(L' x BF) = gi(L' X BF) = L' X BF,
and hd(L' x SF) = gd(L' X SF) = ld. But h21 0 h l (Sd = S2. The properties of
h21 0 hI contradict Proposition 3.14(b). SO IPE,F(S)I = 1.
Let S* E Ro(X) be such that S = S· n L, and let (y, z) E S. Let r > 0 be such
that X n (BE(y,r) X BF(z,r)) <;;;; S·. It thus follows that
Tn (BE(y, r) x BF(z, r)) = {v} X BF(z, r). 0

If (U, V) is a usable pair, then (L(V; U), C(V; U)) is rather well-behaved. We
shall indicate this situation later, but for the time being we restrict ourselves to the
case that F is finite dimensional. Under this assumption we obtain the following
conclusion.

Corollary 3.19 Suppose that F is finite dimensional. Let (U, V) be a usable pair.
Then (L(V; U), C(Vi U)) E [{ReM, and Dc(L(V; U), C(V; U)) is a dense open sub-
set of L(V; U).
Proof: The proof is trivial. In fact, Dc(L(V; U), C(V; U)) is a manifold based on
F, and C(V; U) contains all the homeomorphisms of L(V; U) which are the identity
outside the range of a chart of L(V; U). 0

We have proved two facts on expressibility that have yet to be translated into
a reconstruction result. These are Proposition 3.15(c),(d) and Lemma 3.18(b). In
order to state this reconstruction result, we define a new structure.

Definition 3.20 (a) Let Usbl(X) d;j {(U, V, x) I (U, V) is a usable pair, and
x E Dc(L(V; U))}. Note that H has a natural action on Usbl(X). Let PI «(U, V, x)) =
U and P2«(U, V, x)) = V.
Locally Moving Croups and Reconstruction Problems 147

(b) Let Mus(X, H) d~ (Ro(X), Usb/eX), H; Pi, P2, ::;Ro(X), 0, Ap),

where Ap describes the action of H on Ro(X) U Usb/eX).
Note every x E X is represented in Mus(X, H) by infinitely many objects. That
is, if x E Dc(L(V; U)) and x E Dc(L(V'; U')), then both (U, V, x) and (U', V', x)
represent x. We shall later see that the fact that two triples represent the same
point is expressible in M us(X, H).

Corollary 3.21 For / = 1,2 let FI = (Xl, ,,)) be thin foliated spaces based on
EI and Fl. Suppose that FI is finite dimensional. Let HI be either Ho(F) or
Hl(F), and cp: Hi ~ H2. Then cp can be extended to an isomorphism between
MUS(Xl,Hl) and MUS(X2,H2).

Proof' We indicate the steps in the proof. All of them are similar to arguments
that have already been presented.
We begin by applying 3.11 and 2.10 to Hi and H2, thus obtaining an isomor-
phism between M(Ro(Xl),Hl) and M(Ro(X2),H 2).
Then we apply 3.15(c) and (d) to capture the set of all C(V; U)'s in
M(Ro(XI), HI).
Next we apply 3.14(c) and 2.10 to the set of C(V; U)'s captured above. By
this we capture the set of M(Ro(L(V; U», C(V; U»'s, for all U E Ro(X) and
V E INV(U) - {0}.
The set of all usable M(Ro(L(V; U», C(V; U»'s is then captured by using 3.18.
Finally we apply 3.19 and 3.4 to construct an isomorphism between
Mus(Xl,Hl) and Mus(X2,H2). 0

There are still three easy steps before the proof of Theorem 3.10 can be con-
cluded.

Proposition 3.22 For every x E X there is a usable pair (U, V) such that x E
Dc(L(V; U».

Proof' Let x EX. Let I E r be such that x E B"(' We assume that I = Id.
Since F is thin, there is a thin Y' ~ BE such that B, = Y' X BF. Suppose that
x = (y, z). Let U' E Nbr (y) and V' E Ro(Y'HU' be such that bdu (V') = {y}.
y' ,

Let U = U' X BF and V = V' X BF. Then (U, V) is a usable pair, and x E L(V; U).
o

Proposition 3.23 (aJ There is a first order formula 1/J.(u, W) in the language of
Mus(X, H) such that for every (U, V, x) E Usbl(X, H) and
WE Ro(X): Mus(X, H) F 1/J.[(U, V, x), W] iff x E W.
(bJ There is a first order formula 1/JEq(U, v) in the language of Mus(X, H) such
that for every (U, V, x), (U', V', x') E Usb/eX, H):
Mus(X, H) F 1/JEq[(U, V, x), (U', V', x')] iff x = x'.
148 Matatyahu Rubin

de!
Proof (a) Let 1/;e(u, W) == (3g E Sp H(W) n Sp H(pl(U)))(9(u) -::j::. u).
It is easy to check that 1/;e (u, W) is as required.
de!
(b) Let 1/;Eq(U,V) == (VW E Ro(X))(1/;e(u, W) ...... 1/;e(v' W)). It is trivial that
1/;Eq(U, v) is as required. 0

Proof of Theorem 3.10: We prove the special case when F is finite dimensional.
Combine Corollary 3.21, Proposition 3.22 and Proposition 3.23. 0

Proof of Theorem 3.8: The set

{( T.c, H i ( T.c)) I £ is a parallel family and i E {O, I}} is contained in the class
{(X, Hi (:F)) I :F (X, -) is a thin foliated space based on some E, F, and
F is finite dimensional, and i E {O, I}}. 0

We shall next indicate how to prove Theorem 3.10 without the extra assumption
that F is finite dimensional.

Definition 3.24 Let F be a normed vector space over IR and X be a topological

space.
(a) Let, : fjF -+ X be a map with the following properties: (1), : fjF ~ Rng (,);
(2) ,U!J F ) is closed in X; and (3) ,(BF) is open in X. Then, is called a normal
F-map.
(b) Let r be a set of normal F-maps such that U{Rng(,) I, E r} is a dense
subset of X. Then we say that (X, r) is a regionally normed space based on F with
atlas r.
(c) Let (X, r) be a regionally normed space based on F, and H S H(X). We
say that H is adequate, if {(I 0 h 0 ,-I) U (Idl(X - RngCl))) I h E H(fjF)} ~ H.
We say that (X, H) is an adequate pair ifthere is r such that (X, r) is a region-
ally normed space, and H is an adequate subgroup of H(X).

Proposition 3.25 If (U, V) is usable, then (L(V; U), C(V; U)) is adequate.

Proof: The trivial proof is left to the reader. o

In [12] Theorem 3.43, it was proved that for the class of adequate (X, H) 's,
Dc(X, H) can be captured from M(Ro(X), H). If in the proof of Theorem 3.10 we
apply Theorem 3.43 of [12] rather than Theorem 3.4, then the finite dimensionality
of F is never used.

4 Linear Orders and Circular Orders

In this section we shall demonstrate how the Reconstruction Theorem for Local
Movement Systems can be applied to linear orders and circular orders. We shall
not present the strongest known results, but rather try to give a unified treatment
of the full automorphism groups of such structures. We shall deal with four closely
Locally Moving Groups and Reconstruction Problems 149

related types of structures: linear orders, circular orders, so called "Equal Direction
Structures" and "Equal Orientation Structures" .
The main result in this section is stated in Theorem 4.6. It more or less says
that under certain homogeneity assumptions on such structures, their Dedekind
completion can be reconstructed from their automorphism group.
The interpretation uses only first order formulas. As a result of this it becomes
possible in many cases to distinguish between such automorphism groups by first
order sentences.
The results presented in this section are known. Some of the history of these
results is described in the introduction. However, I am not aware of any published
presentation of a first order reconstruction for the class of circular orders.
Definition 4.1 Let (L, <) be a dense linearly ordered set.
(a) Assume that (L, <) does not have a maximum. The circular order based on
(L, <) is defined as CR «(L, <) d~ (L, Cr), where
Cr = {(x, y, z) E L31 (x < Y < z) V (z < x < y) V (Y < z < xn.
We next define structures associated with (L, <) and CR «(L, <) whose auto-
morphism groups contain respectively, the order reversing permutations of (L, <)
and the orientation reversing permutations of CR«(L, <). Let x E L n denote
(XI, ... ,X n ).
(b) We define the "Equal Direction Structure" based on (L, <) as
ED«(L, <) d~ (L, Ed), where
Ed = {x'ylx,yE L2 and «Xl < x2)A(YI < Y2»V«XI > x2)A(YI > Y2)n. Here
x ' y denotes the concatenation of x and y.
(c) We define the "Equal Orientation Structure" based on (L, <) as
EO«(L,<) d~ (L,Eo), where Eo = {x'ylx,YE L3 and
« Cr(xI, X2, X3) A Cr(YI, Y2, Y3» V (Cr(X2' Xl, X3) A Cr(Y2, YI, Y3»n.
Note that EO( (L, <) is defined only for L's that do not have a maximum.
When needed, we shall denote Cr, Ed and Eo by Cr«(L, <), Ed«(L, <) and
Eo «(L, <) respectively.
Structures of type (L, <) and ED«(L, <) in which L has no maximum and no
minimum, and structures of the types defined in parts (a) and (c) are called almost
ordered structures. We also say that they are based on (L, <). We shall use Lalone
to stand for (L, <).
(d) Let N be an almost ordered structure. We define NS and N r , the strict
and relaxed versions of N as follows. If N = L or N = ED(L), then NS = Land
Nr = ED(L). If N = CR(L) or N = EO(L), then NS = CR(L) and Nr = EO(L).
(e) Let (L, <) be a linearly ordered set. We denote the reverse ordering of < by
<*, and (L, <)* denotes (L, <*). We abbreviate (L, <)* by L*.
Exrunple: Let C denote the circle. Let Cw = {x E C 3 I Xl, X2, X3 are
clockwise oriented}. Then (C, Cw) is isomorphic to CR«([O,1),<). However,
(C, Cw) is not isomorphic to CR «(JR, <).
150 Matatyahu Rubin

On the other hand, CR(Q) == CR(Q ~o).

We denote (L, <) by L, and from now on we assume that L is a dense linear
ordering. If L has a minimum, then min (L) denotes that minimum. If L has a
maximum, then max(L) denotes that maximum.

Definition 4.2 Let L be a linearly ordered set.

(a) We say that Land ED(L) are adequately homogeneous (or for short, we say
that they are adequate) if for every X, y E L2: if Xl < X2 and YJ < Y2, then there is
9 E Aut(L) such that g(x) = y.
(b) We say that CR(L) and EO(L) are adequately homogeneous (or for short, we
say that they are adequate) iffor every X, y E L3: if Cr(x1, X2, X3) and Cr(Yl, Y2, Y3),
then there is 9 E Aut(CR(L)) such that g(i) = fj.

Our definitions are set in such a way that every adequate structure is almost
ordered.

Definition 4.3 (a) For distinct X, Y E L, let (x, y) = {z ELI X < z < V}, (x, (0) =
{z E To I X < z}, (-oo,x) = {z ELI z < x} and (x,y)Cr = {z ELI Cr(x,z,y)}.
Nonempty sets of the first three types are called intervals of L, nonempty sets of
the second and third types are called rays of L, ;md nonempty sets of the forth type
are called circular inter'vals of L.
(b) Let T(L) be the topology on L generated by all intervals of L, and TCr(L)
be the topology on L generated by all circular intervals of L. If N = (L, <) or
N = FD (L), then the set of N -inter'vats is the set of intervals of L, and the set
of proper N -intervals is the set of N-intervals which are not rays. Similarly, if
N = GR(L) or N = EO(L), then the set of N-internls is the set of circular
intervals of L, and every N -interval is a proper N -interval.
If N = (L, <) or N = ED(L), then TN d:J T(L). If N = GR(L) or N = EO(L),
then TN d::j TCR(L). We shall regard N as a topological space equipping it with
the topology TN.
Note that if N is almost ordered, then every proper open subset U of N can
be uniquely represented as the union of a set U of pairwise disjoint N -intervals.
Also, U E Ro(N) iff no two distinct members of U have a common end point. In
particular, every N-interval is a member of Ro(N) - {O, I}.
We next set the conventions for the Dedekind completion of an almost ordered
structure. Recall that we assume that L is a dense linear ordering. _
(c) Let L denote the full Dedekind completion of L. That is, L includes a
minimum and a maximum. Let L denote L - {min(L), max(I)}, and f}-,'r denote
I - {max(r)}.
(i) If N = (L, <), then N is L.
(ii) If N = ED(L), then f..1 d!j ED(L).
Assume that L has no maximum.
Locally Moving Groups and Reconstruction Problems 151

(iii) If N = CR(L), then N d;j CR(LCr).

(iv) If N = EO(L),then N d;j Eo(LCr).
Recall that if M = (A, ... ) is a structure, then A is denoted by IMI.
Note that if N is almost ordered, then N is almost ordered.
Proposition 4.4 (a) Aut(L) :S Aut(ED(L» :S H«(L, T(L))). Also,
Aut(ED(L» = Aut(L) U {g I9 : L ~ L*}, and the index of Aut(L) in Aut(ED(L»
is < 2.
-(b) Aut( CR(L» :S Aut(EO(L» :S H«(L, TCT(L»)). Also,
Aut(EO(L» = Aut( CR(L» U {g I 9 : CR(L) ~ CR(L*)}, and the index of
Aut(EO(L» in Aut(CR(L» is:S 2.
(c) Let N be an almost ordered structure. For every 9 E Aut(N) there is a
unique g E Aut(N) such that g extends g. Let Aur(N) = {g I 9 E Aut(N)}.
The function ,N defined by:

U f-+ int N (d N (U», U E Ro(N); 9 f-+ g, 9 E Aut(N)

is an isomorphism between M(Ro(N), Aut(N» and M(Ro(N), Aur(N».
We identify Aut(N) with AunN).
(d) Let N be an adequate structure of the form (L,<) or CR(L). Let x,fjE INln
be such that: (i) if N = (L, <), then for every 1 :S i < j :S n, Xi < Xj and Yi < Yj;
(ii) if N =
CR(L), then for every 1 :S i < j < k :S n, (Xi, Xj, Xk), (Yi, Yj, Yk) E Cr.
Then there is 9 E Aut(N) such that for every i =
1 ... n, g(xd Vi.=
Ifin addition, u,v E INI and x,fjE (u,v)N, then such a 9 exists in
Sp Aut(N)«U, v)N).
(e) If N is adequate, then (Ro(N), Aut(N» is a local movement system.
Proof: Trivial. o
Remark: Part (d) in the above proposition is the only fact that we need to use
in order to construct automorphisms. It will be used at several points in the proof
of the main theorem, not always with an explicit reference.
Corollary 4.5 For I = 1, 2 let Nr be an adequate almost ordered structure. Suppose
that rp : Aut(Nd :::: Aut(N2 ). Then there is
1/J : M(Ro(Nd, Aut(Nd) ~ M(Ro(N2 ), Aut(N2 » such that 1/J extends rp.
We now state the main result of this section.
Theorem 4.6 For I = 1,2 let N r be an adequate almost ordered structure.
(a) Suppose that rp: Aut(Nd :::: Aut(N2 ). Then there is a : Nf:::: N!j such that
a induces rp.
(b) Suppose that Aut(Nd and Aut(N2 ) are elementarily equivalent. (That is,
every first order sentence that holds in one of them, holds in the other.) Then the
structures Act(N[, Aut(Nd) and Act(Nf2, Aut(N2 »
are also elementarily equiva-
lent.
152 Mataty ahu Rubin

use Theore m 3.1

Remar k: Note that N is a locally compac t space. So we can
. Howeve r, the proof
as the main step in the proof of part (a) in the above theorem
order objects to represe nt
of Theore m 3.1 uses second order formul as and second
result by using only first
points. We shall obtain a stronge r type of a reconst ruction
strengt h is the conten t of part
order formula s and first order objects . This extra
(b) of the theorem .

Lemm a 4.7 Let N be adequate. Then

Aut(N S) = {YlY2 I Yl, Y2 E Aut(N) and fix(yt), fix(Y2) i- O}.

Proof: Let N = (L, R) and NS = (L, RS). Let

G = {YlY21 Yl, Y2 E Aut(N ) and fix(yI), fix(Y2) i- O}.

S to prove.
We prove that G ~ Aut(N S). If N = N , then there is nothing
that fix(y) i- O. So there is x E RS
Suppos e otherw ise. Let Y E Aut(N) be such
such that y(x) = x. Let y E RS Then x' y E R. So y(x) . y(Y) E R. That is,
S
X' Y(il) E R. Hence y(Y) E R . Hence Y E Aut(NS ).
S E Aut(N ') such
We next prove that for every Y E Aut(N ) there are YI,Y2
S = (L, <). Let x be
that Y = YlY2 and fix(YJ),fix(Y2) i- O. Suppos e first that N
y(x) > x. Let y> y2(X). Then there is
such that y(x) i- x. We may assume that
Y] E Aut(L) such that yd(y, 00) = Id and yd(x, y(x» = y[(x, y(x». Let Y2 = Ylly.
Hence YlY2 = Y and Y2[(X, y(x» = Id.
Suppos e next that NS = CR(L). There is an interva l (x,
y)C,· such that
y((x, y)Cr) n (x, y)Cr = 0. Let 1l = y(x) and v = y(y). Hence 1l, v ~ (x, y)Cr. We
easily that y, E (x, v)Cr and that y E (x, 1l )Cr.
may assume that v i- x. It follows 1l

Let w, z E (v, x)Cr be such that (v, w, z) E Cr.

It. is easy to constru ct Yl E Aut(NS ) such that Yl [(x, y)Cr =
y[(x, y)Cr,
Yl((Z, W)CT) = (z, w)Cr and yd(w, z)Cr = [d. Let Yz = ally. Then
(ll, V)CT ~ fiX(Y2), and Y = YIY2.
0

Coroll ary 4.8 There is a first order formula <fs(Y) such that
for every adequate N
and Y E Aut(N) : M(Ro( N), Aut(N» 1= <fsly] iff Y E Aut(N S ).
o
Proof: This follows triviall y from the previou s lemma.

Let us now see that the set of proper interva ls of N is definab le

in
M(Ro( N), Aut(N» by a first order formula .
Let B he a comple te atomle ss BA and G :S Aut (B). Let a E
B. We say that
b, c :S a there is Y E Sp G (a) such that
a is flexible in (B, G), if for every nonzero
y(b),c i-O.
of M (B, G) that
Note that there is a first order formul a r.p Fix (x) in the languag e
express es the fact that x is flexible in (H, G).
Locally Moving Groups and Reconstruction Problems 153

Proposition 4.9 Let N be an adequate almost ordered structure and U E Ro(N)-

{O,1}. Then:
(aJ U is an N-interval iff U is flexible in (Ro(N), Aut(N)).
de! -
(bJ The formula </'Invl(U) == (U E Ro(N) - {a, I}) /\ </'Flx(U)
defines the set of N -intervals in every adequate almost ordered structure N.
(cJ The formula
</'Pinvl(U) d!1 </'Invl(U) /\ (3g E Aut(NS))((U I g(U)) /\ (g(U) I U))
defines the set of proper intervals in every adequate almost ordered structure N.

Proof: The proof is trivial. o

There is now a choice of ways of representing the points of N by objects from

M(Ro(N), Aut(N)). We shall represent these points by pairs of N-intervals with
a common endpoint. This representation has the advantage of working for all four
types of almost ordered structures using exactly the same formulas.

Definition 4.10 Let N be an adequate almost ordered structure. We shall use a

simplified notation for N-intervals. Let x, yEN be distinct.
(i) If N is of the form (L,<) or CR(L), then (x,y) denotes respectively {z E
L I x < z < y} and {z E N I (x, z, y) E Cr(LCr)}.
(ii) If for some L, N = ED(L), then choose LN E {L, L*}, and define (x, y) to
be {z E LN I x < z < y}.
(iii) If for some L, N = EO(L), then choose LN E {L, L*} such that N =
EO(LN). We may assume that N = Eo(LCr). Now, define (x, y) to be
{z EN I (x, z, y) E Cr((LN)Cr)}.
Note that in (ii) and (iii) one needs to assume the existence of a global choice
function. But we ignore this little inaccuracy.
Let x E INI and U, V be proper N-intervals. We say that (U, V) represents x
if there are distinct y, zEN - {x} such that either U = (x, y) and V = (x, z) or
U = (y,x) and V = (z,x).
A pair (U, V) of proper N -intervals is called a representative if it represents some
x EN; x is then denoted by xu,v. Let Rep (N) denote the set of representatives in
N.
Let EqrN be the following relation on Rep(N).

Eqr
N
= {(U, V)
~ ~
E (Rep(N))
2
I Xu = xv}.
LeIllIlla 4.11 (aJ There is a first order formula </'Rep(U, V) in the language of
M(Ro(N), Aut(N)) such that for every adequate N and every U, V E Ro(N):
M(Ro(N), Aut(N)) f= </'Rep[U, V] iff W, V) E Rep(N).
(b J There is a first order formula </' Eqr (0, V) in the language of
M(Ro(N), Aut(N)) such that for every adequate N and every 0, V E Rep(N):
M(Ro(N),Aut(N)) f= </'Eqr[O, V] iff Xu = xv,
154 Matatyahu Rubin

Proof: Let _Ro denote the difference operation in the Boolean algebra Ro(N).
def
(a) Let 'PRep(U, V) == 'PPinvl(U) /\ 'PPinvl(V)/\
«(U < V) /\ 'PPinvl(V _Ro U)) V «V < U) /\ 'PPinvl(U _Ro V))).
It is easy to see that 'PRep(U, V) is as required.

Rep(N). Then °
(b) We define the "same side equivalence" of representatives. Let 0, V E
Sseq V, if there are X,UI,U2,VI,V2 E INI such that
either for every i E {1,2}, Ui = (Uj,x) and Vi = (Vi,X),
or for every i E {I, 2}, Ui = (x, Ui) and Vi = (x, Vi).
Let 'PSseq (U, V) == l\i,jE{I,2}«(Ui = Vj) V 'PRep(Ui, Vj)).
- - def

Claim 1: Let N be adequate and 0, V E Rep(N). Then U Sseq V iff

M(Ro (N), Aut(N)) F= 'PSseq [0, V].
°
Proof of Claim 1: It is trivial that if Sseq V, then
M(Ro(N), Aut(N)) F= 'PSseq [0, V].
Let 0, if E Rep(N), and assume that M(Ro(N), Aut(N)) F= 'PSseq [0, V]. Then
UI and U2 are comparable under inclusion and have precisely one common endpoint
xu1,u,. Also, VI is comparable under inclusion with each Ui and has at least one
endpoint in common with each. So VI must have xu1,u, as an endpoint and must
lie on the same side of XU 1,u2 as do UI and U2 • This applies likewise to V2, proving
Claim 1.
The equivalences classes of Sseq represent every x E N twice.
Both (YI,X), (Y2,x))/Sseq and (x,zI), (X,Z2))/Sseq represent x. Our next goal is
to find a formula that expresses the fact that two representatives represent the same
point.
-+ - de] -+ -+ -+-+

Let 'PEqr(U, V) == (Vg E Aut(N))('Psseq(U, g(U)) -> 'PSseq (V, g(V))).

If N is adequate, then for every distinct x, yEN there is 9 E Aut (N) such that
g(x) = x and g(y) f:. y. It follows easily from this fact that 'PEqr is as required. 0

It now remains to show that the relation of N r (which by Definition 4.1(d) is

either Ed or Eo) is definable by a first order formula in M(Ro(N), Aut(N)).
For a linear ordering L let Edt(L) d2 {£'y I x,y E L3 are 1-1 sequences,
and the set {(Xi, Yi) I i = 1,2, 3} is an order preserving function}.
Note that Edt (L) is first order definable in ED (L), and Ed (L) is first order defin-
able in (L, Edt(L)). So it does not matter whether we capture Ed or we capture
Edt. For an almost ordered N based on L, let the "relaxed uniformized relation" of
N be defined as follows: if N = (L, <) or N = ED(L), then Ru(N) = Edt(L); and
if N = CR(L) or N = EO(L), then Ru(N) = Eo(L). Let the "relaxed uniformized
version" of N be defined as follows: N ru d2 (INI, Ru(N)).
Proposition 4.12 (aJ There is a first order formula 'P.(O, V) in the language of
M(Ro(N), Aut(N)) such that for every adequate N, (j E Rep(N) and V E Ro(N),
Locally Moving Groups and Reconstruction Problems 155

M(Ro(N), Aut(N)) F 'P,,[O, V] iff Xu E V.

(b) There IS a first order formula 'PRu(01,02,03,i71 ,V 2 ,V3) zn
the language of M(Ro(N), Aut(N)) such that for every adequate Nand
0 1,0 2,03, VI, V2, V3 E Rep(N):
M(Ro(N), Aut(N)) F 'PRu[Ol, O2 , 0 3, VI, V2 , V3] iff
RU(XUI,XU2,XU3,XVI,XV2'XV) holds in N ru .

Proof: Part (a) is needed in the proof of (b).

. - de! - - .
(a) It IS easy to see that 'P,,(U, V) == (::3g E SPAut(N)(V))--''PEqr(9(U), U) IS as
required.
(b) We denote members of Rep (N) by the letters 1', s. For I = 1,2 let i1 denote
a member of (Rep (N))3. Let

/\
(::39 E Aut(JVS))(:lrE (Rep(N))3) /\ ('P,,(ri, W1,;) 1\ 'P,,(g(r;), W 2,i)))).
o::;i<j9
It follows easily from Proposition 4.4(d) that 'PRu is as required. 0

Proof of Theorem 4.6: (a) The proof of Part (a) follows the pattern of the proof
of the Reconstruction Theorem for Local Movement Systems. So we just indicate
the main steps in the proof.
For an adequate N, let M = Mru(N) be defined as follows.
M = (IMI, ::;Ro, 0, Ap, Ru(N), E), where IMI = IM(Ro(N), Aut(N))1 u INI, and
0, Ap and E are defined as follows. ° is the composition operation on Aut(N),
(defined in a trivial way outside (A ut (N) )2); Ap is the application function defined
on A ut( N) x (INI U Ro (N)); and E is the belonging relation defined on INI x Ro (N).
By Propositions 4.11 and 4.12, iffor 1= 1,2, N/ is adequate, then every isomor-
phism between M(Ro(NJ), Aut(Nl )) and M(Ro(N2), Aut(N2 )) can be extended to
an isomorphism between Mru(NJ) and Mru(N2)'
Combining this with Corollary 4 ..5, we conclude that if for I = 1,2, N/ is ade-
quate, then for every 'P : Aut(Nt) ~ Aut(N2) there is
f3 : Mru(NJ) ~ Mru(N2) such that /3 extends 'P. Let Cl' = f3[N 1. Then Cl' : Nfu ~ Nfl!,
and Cl' induces 'P. By the mutual definability of Ed and Edt, Ct : Nf ~ Nf. This
proves Part (a).
(b) Let N be adequate. We define the set of "automorphism representatives" of
N. Repa(N) d~ {(9l,92) E (Aut(N))2 I (var(91), var(92)) E Rep(N)}.
For (91,92) E Repa(N), let. x 91 ,9z =
de!
x var(gJ),var(g2)'
Let Eqra(N) d~ {(1'1, 1'2) E (Repa(N))21 X"I = x r ,).
By 4.11 and 1.19(b), there are formulas 'PRepa(X1,X2) and 'PEqra(Xl,X2) in the
156 Matatyahu Rubin

language of groups such that for every adequate N, 'P Repa defines in Aut (N) the
set Repa(N), and 'PEqra(Xl,X2) defines in Aut(N) the relation Eqra(N).
Similarly, by 4.12, 4.11(b) and] .19(b), there are formulas 'PRua and 'PApa in the
language of groups, that respectively define in Aut (N) the relation
{f E (Repa(N»61(Xtll"" Xt6) E Ru(N)}, and the application function on Aut(N)x
(Repa (N)/ Eqra (N»).
It now follows easily from the existence of the above formulas that for every for-
mula 'P(Xl, ... , Xk, hI, ... , h m ) in the language of Act(N ru , Aut (N») there is a for-
mula 'P* (II, gl, ... , /k, gk, hI, ... , h m ) in the language of groups such that for every
adequate N, (h,gl), ... (/k,gk) E Repa(N) and
hI,.· .,h m E Aut(N): Act(Nru,Aut(N) F= 'P[xh,91'" .,xj.,9k,hl , ... ,h m ] iff
Aut(N) F= 'P*[h, gl,· .. , /k, gk, hI, . .. , h m ].
It follows that for every adequate NI and N 2 : if Aut(Nd == Aut(N2)' then
Act(N)'u, Aut(N]) == Act (!\T2" , Aut(N2». By the mutual definability of Ed and
Edt, this implies that Act(Nf, Aut(Nd) == Act(N21 Aut(N2». 0

Theorem 4.6 is a powerful tool in proving that the automorphism groups of

some adequate almost ordered structures can be characterized in the class of all
such automorphism groups by a first order sentence. The following theorem due to
Gurevich and Holland [4] is such an example.

Corollary 4.13 There are first order sentences 'PIR, 'PilJ in the language of groups
such that for every adequate N: Aut(N) F= 'Pm iff N == IR and Aut(N) F= 'PilJ iff
N == Q OT' N - N == Q.
Proof: The proof is not difficult. It uses the fact that the separability of N,
the completeness of N and the countability of an orbit in Act(Nr, Aut(N» can be
expressed by first order sentences in the language of Act(Nr, Aut(N».
o

References

[1] J. L. Alperin, J. Covington and D. Macpherson, Automorphisms of quotients

of symmetric gmups, this volume.
[2] R. Bieri and R. Strebel, On groups of PL-homeomorphisms of the real line,
Preprint 1985.
[:\] J. R. Riichi, D1I: Roolesche Partialordnunq und dip jJaanmg von Gefiigen, Por-
tugal. Math. 7(1948), 119 190.
[4] Y. Gurevich and W. C. Holland, Recognizing the real line, Trans. American
Math. Soc. 265(1981),527-.534.
[5] M. Giraudct and .J. K. Truss, On distmguishing quo/zents of ordered permuta-
tion groups, Quat. J. Oxford 45(1994),181-209.
Locally Moving Groups and Reconstruction Problems 157

[6] W. C. Holland, Transitive lattice-ordered permutation groups, Math. Zeit.

87(1965), 420-433.

[7] S. Koppelberg, Handbook of Boolean Algebras, Edited by J. D. Monk, Vol 1,

North Holland, Amsterdam, 1989.

[8] W. Ling, A classification theorem for manifold automorphism groups, Preprint

1980.
[9] D. Maharam, On homogeneous measure algebras, Proc. Nat. Acad. Sc.
28(1942),108-111.
[10] S. H. McCleary, Groups of homeomorphisms with manageable automorphism
groups, Comm. in Algebra 6(1978), 497-528.

[11] E. B. Rabinovich, On linearly ordered sets with 2-transitive groups of automor-

phisms, Vesti Akad. Nauk Belaruskai SSR 6(1975), 10-17.

[12] M. Rubin, On the reconstruction of topological spaces from their groups of

homeomorphisms, Trans. American Math. Soc. 312(1989), 487-538.

[13] M. Rubin, On the reconstruction of Boolean algebras from their automorphism

groups, in Handbook of Boolean Algebras, ed. J. D. Monk, Vol 2 Chapter 15,
547-605, North Holland, Amsterdam, 1989.

[14] M. Rubin, On the automorphism groups of homogeneous and saturated Boolean

algebras, Algebra Universalis 9(1979), 54-86.

[15] M. Rubin, The reconstruction of trees from their automorphism groups, Con-
temporary Mathematics 151(1993), American Math. Soc.
[16] M. Rubin and Y. Yomdin, in preparation.

[17] S. Shelah and J. K. Truss, On distinguishing quotients of symmetric groups, in

preparation.

[18] F. Takens, Characterization of a differentiable structure by its group of diffeo-

morphisms, Bol. Soc. Bras. Mat. 10(1979), 17-26.

[19] J. K. Truss, On recovering structures from their automorphism groups, this

volume.
[20] J. V. Whittaker, On isomorphic groups and homeomorphic spaces, Ann. Math.
78(1963),74-91.
 Infinite Jordan Permutation Groups

S. A. Adeleke

University of Western Illinois

Macomb, Illinois, 61455
USA

1 Introduction

A Jordan group is a transitive permutation group which contains in a non-trivial

way a smaller permutation group, the smaller group fixing every point outside the
smaller space. More precisely, a transitive permutation group (G, 0) is Jordan if
(i) G contains a group H, (ii) 0 contains a non- empty, non-singleton set r, (iii) H
fixes every point in O\r with (H, r) transitive and, (iv) in case 10\rl is finite, then
G is not (IO\rl + I)-transitive on O.
This paper presents a summary of the extensive work on invariant relations in
infinite primitive Jordan groups; it indicates some articles close to that work, and
gives some new results on the topic. What makes Jordan groups very interesting
is that with very few assumptions, one can get far-reaching conclusions as to their
invariant relations. The relations are: linear order, linear betweenness relation,
circular order, separation relation, semi linear order (that is, the order on a "tree"),
C-relation (that is, the relation on maximal chains on a "tree"), D-relation (relation
on the "leaves" or points at infinity on a "tree"), Steiner systems, and limits of some
of these relations. Those are all the possibilities, according to the work in [1]-[4],
[6], [7], [ll], [24]-[27], [31]. The contents below also includes some new results on
Jordan sets and linear order, and on invariant relations on a group satisfying slightly
weaker properties than those of Jordan groups.
One new result below contains as a corollary that a transitive primitive or-
dered permutation group need not be weakly doubly transitive, the right regular
representation of a subgroup of ffi. or periodic. The second new result shows that
if a primitive permutation group has a syzygetic subset, then it has an invariant
semilinear order relation, a B-relation, a C-relation or aD-relation.
This summary is limited in several ways. It is limited to invariant relations;
so, only a brief mention is given in section 5.8 to the very important work on
normal subgroups and other structures of Jordan groups contained in publications
like [13H15], [17], [26], [28], [29], [30]. It is limited to infinite Jordan groups; its

159
W. C. Holland (ed.), Ordered Groups and Infinite Permutation Groups, 159-194.
© 1996 Kluwer Academic Publishers.
160 S. A. Adeleke

only reference to finite Jordan groups is to areas of very close connection to the
infinite case. Thus, the seminal work of Jordan [19} and the pioneering work in [20],
[21] are merely mentioned. Even on invariant relations, the paper does not give
details of the work in [11], [16], and [32] on when an invariant Steiner system is a
projective or affine space. Some survey articles like [27] and [20] address some of
the areas of limitation of this article.
There is also an excellent survey on Jordan groups by Macpherson [22]. But its
emphasis is different from the one in this paper. The survey in [22] highlights the
pleasant interplay of the works on infinite Jordan groups and Model Theory. This
paper, however, leaves out Model Theory altogether and focuses on infinite Jordan
groups, especially on the evolution of the ideas that led to the final classification
of the invariant relations on Jordan groups. Those early stages on bounded groups
(generalization of finitary groups) and the rapid sequence ofresults in the classifica-
tion of infinite Jordan groups with proper primitive Jordan sets get most attention
here. Whenever possible, it presents alternative but elementary proofs. The overall
goal is that this article, when read in conjunction with the very short summary in
[4, section 4] on the later proofs on Jordan groups not containing proper primitive
Jordan sets should give useful insight to the reader.
A word about proofs in infinite Jordan groups. At the heart of almost all the
proofs are maximal Jordan sets containing one element and excluding others. The
reasons are two: First is the fact that the union of two nondisjoint Jordan sets is
Jordan, the union of a chain of Jordan sets is Jordan and the image of a Jordan
set under any group element is also a Jordan set. Second is that the primitivity of
the group (0,0) can be used to produce from one Jordan set a collection of larger
Jordan sets. For example, in the case of a primitive but not 2-transitive Jordan
group (0,0) with a Jordan set r, a E l' and /3 E 0\1', the maximal Jordan set
containing a and excluding /3 exists, is unique, and with all other such maximal
Jordan sets plays a role in revealing the invariant relations in (0,0). Details about
these general features of the proofs are in section 4 below.
The contents is divided into sections as follows: Section 2: definitions and
notation; section 3: statement of the classification theorem; section 4: vital lemmas
and properties; section 5: bounded permutation groups and comments on other
works; section 6: Jordan groups with proper primitive Jordan sets; section 7: general
simply primitive Jordan groups; sections 8 and 9: new results.

2 Definitions and Notation

2.1 Definitions on Permutation Groups

1. A permutation group (0,0) is transitive if for every pair a and /3 in 0,
there is a permutation in 0 mapping (j; to /3.
2. The group (G,O) is n-transitivc if for every pair (al,a2, ... ,a n ) and
(/31, (3'2, ... , (3,,) of n-tuples with distinct components in
0, there is a permutation
Infinite Jordan Permutation Groups 161

in C mapping (0'1,0'2, ... , O'n) to (fJ1 , fJ2, ... , fJn). If (C, 0) is n-transitive for every
natural number n, then it is said to be highly transitIVe.

3. In a permutation group (C, 0), a non-empty subset r ~ 0 is a block if for

every 9 E C, either fg n r = ¢; or rg = r. A permutation group (C,O) with no
blocks besides 0 and singleton sets is said to be primitive. A permutation group
(C, 0) is n- primitive with n 2: 2 if for any distinct n - 1 elements 0' 1, 0'2, ... , O'n-l
in 0, the group (H, 0\ {O'l, 0'2, ... , O'n- d) is primitive where H is the collection of
all 9 E C such that g fixes each of 0'1, 0'2,"" O'n-l. A primitive group which is not
2-transitive is called simply primitive.

4. Let (C, 0) be a transitive permutation group. A non-empty, non-singleton

subset, r, of 0 is a lordan set if for every 0' and fJ in r, there is a permutation in
C mapping 0' to fJ and fixing every point in O\f. The Jordan set f is a primitive
Jordan set if the action on r of the subgroup of C fixing every point of O\f is
primitive. In a similar way, n-transitive and n-primitive Jordan sets are defined.

5. The support of a permutation h, denoted supp(h), is the set of elements

moved by h. The support of a subgroup H of permutations, denoted supp(H), IS
the union of the supports of the permutations in H.

6. As usual, /f/ shall represent the size of set f.

7. A Jordan set f in (C, 0) is a proper lordan set if O\f is infinite, or if finite,
C is not (IO\fl + I)-transitive. Thus a Jordan set is proper if its existence cannot
be derived trivially from multiple transitivity of (C, 0). Proper primitive Jordan
sets and proper n-transitive Jordan sets have similar definitions.

8. A transitive group (G, 0) is a lordan group if 0 contains proper Jordan sets.

9. In a permutation group (C,O), if (J" E 0, then C<7 denotes the subgroup of

C fixing (J". Similarly, if (J"l, (J"2,.··, (J"n EO, then C U ,<7,.'(7 n denotes the subgroup of
C fixing each of (J"1, (J"2, . . . , and (J"n. In the same way, if II ~ 0, Cell) denotes the
subgroup of C fixing each element in II. Two Jordan sets El and E2 are typical if
El n E2 -# ¢;, El \E2 -# ¢;, and E2 \E 1 -# ¢;.
10. In a Jordan group with Jordan set E, every set of the form Eg where 9 E C
is called a C-translate of E, or a translate of E under C.

11. In any group under consideration, M P 1(;31, fJ2, ... ,fJk; 0') shall denote the
maximal primitive lordan set containing 0' and excluding fJl, fJ2, ... , fJk whenever
the set exists. Likewise M J(fJ1, fJ2, ... , fJk; 0') shall mean the maximal lordan set
containing 0' and excluding fJ1, fJ2, ... , fJk.

12. Throughout, A C B means set A is a proper subset of set B.

162 S. A. Adeleke

2.2 Definitions of Structures

l. Let (n,::;) be a dense linear order. Define on n:
pea; (3, ,) if and only if (3::; a ::;, or ,::; a ::; (3.

Then p is a dense linear betweenness relation induced by the dense linear order.
2. A partial order ::; on a non-empty set E is an upper semilinear order if
(i), every pair in E has an upper bound in E, and (ii), the set of upper bounds of
any element of E is linearly ordered. Then, (E,::;) is called an upper semilinearly
ordered set. (An upper semi linearly ordered set is like a set of points on an inverted
"tree" .)
Let a, (3 be any two incomparable elements in a nonlinear upper semilinearly
ordered set (E, ::;). Define

U {a, (3} := {o' EEl u ~ a /I. u ~ (3}

T{a,(3}:= {u EEl u < wVw E U{a,(3}}.

Then a, (3 E T {a, (3} since a and (3 are chosen to be incomparable. Define"" on

T{a,(3}:
~ "" 17 if and only if (3J.l E T{ a, (3} )(~ ::; J.l /1.17 ::; J.l).

Then it is not difficult to verify that"" is an equivalence relation on T{ a, (3} and

that a, (3 belong to different classes of ~. Define U { a, (3} as a branch point in
(E, ::;), and refer to each class of"" as a branch below the branch point U {a, (3}.
3. A ternary relation B on a non-empty set n is a general betweenness relation
(and if B( a; (3, ,) holds then we shall say that a lies between (3 and, ) if it satisfies
the following conditions:
(B1) B(a; (3,,) --> B(a;" (3)
(B2) B(a; (3,,) /I. B((3; a,,) +--+ a = (3
(B3) B(a;(3,,) --> B(a;(3,6) V B(a;,,6)
(B4) -,B(a; (3,,) --> (36 f- a)(B(6; a, (3) /I. B(6; a,,))
If, in addition
(B5) a f- (3 -+ (3,)(1 f- a /I., f- (3/1. B(,; a, (3))
then we say (n, B) is dense.
The pair (n, B) of a non-empty set n and a general betweenness relation B is
called a general betweenness set.
4. A ternary relation K on a non-empty set n is a C-relation (or a chain
relation) if the following conditions hold:
(Cl) K(a; (3, ,) -+ K(a;" (3)
(C2) K(a;(3,,) ---+ -,K((3;"a)
Infinite Jordan Permutation Groups 163

(C3) (K(a; (3, r) 1\ -,K(8; (3, r)) --+ K(a; 8, r)

(C4) (a 1: (3) --+ (3/L)«J-l 1: (3) 1\ K(a;{3,J-l»
(C5) (Va, 13) (377) (I< (77; a, 13»·
If in addition
K(a; (3, r) --+ (38)(K(8; (3, r) 1\ K(a; (3, 8))
holds, then (n, K) is said to be dense.
(Any set possessing a C-relation can be identified with some collection of maxi-
mal chains of an upper semilinearlyordered set whose union is the whole semilinearly
ordered set.)
5. A quaternary relation D on a non-empty set n is a D-relation if it satisfies
the following conditions:
(D1) D(a, 13; r, 8) --+ (D(j3, a; r, 8) 1\ D(a, 13; 8, r) 1\ D(r, 8; a, 13»)
(D2) D(a,j3;r,8) --+ -,D(a,r;j3,8)
(D3) D(a,j3;r,8) --+ (D(a,j3;8,e) V D(r,8;a,e»
(D4) a, 13, r distinct --+ (38)«r 1: 8) 1\ D(a, (3; r, 8».
If also
(D5) D(a,13; r,8) --+ (3c)(D(e,j3;r, 8) 1\ D(a,e;r,8)
I\D(a, 13; c, 8) 1\ D(a, 13; r, e»
then we say (n, D) is dense. (A set with a D-relation can be identified with a set
of "points at infinity" or "leaves of a tree.")
6. A circular order is a ternary relation p satisfying the following conditions:
(pI) p(a, 13, r) --+ p({3, r, a)
(p2) (p( a, 13, r) 1\ p(j3, a, r» <-> «a = 13) V (13 = r) V (r = a»
(p3) p(a,j3,r) --+ (p(a,{3, 8) V p(8, {3, r»
(p4) pea, 13, r) V p(j3, a, r)

7. Let p be a circular order on a non-empty set A. A group G of permutations

of A preserves the separation relation associated with p on A if and only if every
element of G either preserves or reverses the circular order p. That is,

(Vg E G) [ (Va, 13, r E A)(p(a, (3, r) --+ p(ag, (3g, rg») ]

V(Va, 13, r E A)(p(a, 13, r) --+ p(ag, rg, j3g» .

The separation relation associated with p is the quaternary relation S defined by:

Sea, 13; r, 8) <-> «p(a, 13, r) 1\ pea, 8, (3» V (p(a, r, (3) /\ p(8, a, (3»).

8. Let k be a positive integer greater than 1, and let A be a non-empty set.

A non-trivial Steiner k-system on A is a collection of subsets of A of the same size
greater than k called blocks such that given any k distinct elements in A, there is a
unique block containing them.
164 s. A. Adeleke

9. Let (Q, :::;) be a linearly ordered set. Denote by 0 the Dedekind completion
of Q (see [18, section 1.8] for details). A subset r of 0 is coterminal in 0 iffor every
w E 0, there exist /1, /z E r such that /1 :::; w :::; /z·
10. Let (G, Q) be a transitive permutation group preserving a linear order :::;
on Q. Then, (G, Q) is said to be weakly doubly transitive [18, section 4.3] iffor any
given II < fi < / in Q, there exists g E Go such that fig ~ /, and there exists
hE G-y such that fih :::; ll. The group (G, Q) is said to be periodic if there exists an
order-preserving permutation z of 0 preserving:::; such that
(i) wz > w, Vw EO
(ii) there exists II E Q such that {llZn I n E 2} is coterminal in 0, and
n
(iii) the set of all permutations of that commute with each element of G (that is,
the centraliser of G) is precisely (z), the group generated by z.

3 The Final Classification Theorem

The full proof of the classification theorem is in [2]-[4].

Theorem 3.1 Let (G, Q) be an infinite primitive Jordan group.
If(G,Q) is simply primitive, then G leaves invariant a linear order, a semilin-
ear order or a C -relation on Q.
If (G, Q) is 2-transitive not 2-primitive, then G leaves invariant a linear be-
tweenness relation, a general betweenness relation, a C -relation, a D-relation or a
Steiner system.
If (G, Q) is 2-primitive not 3-transitive, then G preserves a circular order, a
D-relation, a limit of general betweenness relations or of D-relations.
If (G, Q) is 3-transitive but not 3-primitive, then G preserves a separation re-
lation, a D-relation, a Steiner system or a limit of Steiner systems.
If (G, Q) is 3-primitive, then it is highly transitive or it preserves a Steiner
system or a limit of Steiner systems.
Remark 3.2. In sections 5, 6 and 7 below, we set out chronologically the devel-
opment of the proof of the classification theorem. But we omit the sketch proofs
on 2-transitive Jordan groups with no proper primitive Jordan sets for two reasons.
Firstly, [4, section 4] contains very short sketch proofs of the omitted part. Secondly,
the account and sketch proofs below should help the reader with the intuition nec-
essary for a quick understanding of all the full proofs on the topic. A good reference
for definitions of limits of betweenness relations, D-relations and Steiner systems is
[4, Definitions 2.1.9 and 2.l.10]. The limit cases in Theorem 3.1 occur when (G, Q)
contains no proper primitive Jordan sets.

4 Vital Lemmas and Properties

Let (G, Q) be a Jordan group, infinite or finite. Let p denote any of the following
Infinite Jordan Permutation Groups 165

properties of ~ :
~ is a Jordan set;
~ is a primitive Jordan set;
~ is a t-transitive Jordan set for some t E Ii;
~ is a t-primitive Jordan set for some tEN.

The following lemmas are not surprising and are easy to prove (see [2]).
Lemma 4.1.1 If~ is a Jordan set with property p, then ~g is a Jordan set with
property p for any 9 E G.
Lemma 4.1.2 If {~i liE I} is a chain of Jordan sets with property p , then
U{~i liE I} is a Jordan set with property p.

Lemma 4.1.3 Let ~l and ~2 be Jordan sets with property p such that ~l n~2 # cpo
Then ~l U ~2 is a Jordan set with property p.
The next Lemma is a simple consequence of the three Lemmas above as the
sketch proof shows.
Lemma 4.1.4 If (G, 0) is primitive and contains a proper Jordan set with prop-
erty p, then 0 has property p.
Sketch proof. A maximal Jordan set ~ with property p exists by Lemma 4.1.2
and must be O. Otherwise by the primitivity of G, there is 9 E G such that
~g n ~ # cp and ~g\~ # cp, which will imply that ~g U ~ is a Jordan set with
property p strictly larger than ~, a contradiction to maximality of ~.
One of the few results on finite Jordan groups which is extended repeatedly in
the proofs on the infinite case is the following.
Lemma 4.1.5 [27, Lemma 2, section 3] Suppose (G,O) is a finite Jordan group.
Let r 1 be a non-trivial Jordan subset of 0 which is not a block of imprimitivity
for G, let r 2 be minimal amongst Jordan subsets containing r 1 properly, and let
d := r 2 \r 1 .
(aJ Then d is a block of imprimitivity for G{r 2 } in r 2 , and Idl < Irll.
(b) If Idl = 1 then r 1 is an improper Jordan subset for G in O. If Idl > 1
then r 1 is a proper Jordan set and d is a Jordan subset for G{rt} in O\r 1 .
(c) If p is the G{r,}-congruence on r 2 having d as a block of imprimitivity
then G{r 2 } is 2-transitive on rd p; moreover, if r is any Jordan set contained in
r 2 then, provided that r is not contained in a single p-class, r is a union of p-classes
and r/p is a Jordan subset for G{r 2 } in r 2 /p.
4.1.6 Remark on Lemma 4.1.5. (a) The Lemma leads to a chain of Jordan
sets r 1 C r 2 C ... C rk where for each i, ri+l \ri is a block of the unique max-
imal congruence of GCO\ri+d on ri+l, and uri = rk = O. The sets ri+l \ri
are called minimal increments. If (G, 0) is primitive, it follows immediately that
Irk \rk-ll = 1 and thus (G,O) is 2-transitive.
166 S. A. Adeleke

(b) One extension of Lemma 4.1.5 to the infinite case defines r 1 as a maximal
Jordan set with respect to containing one element and excluding a finite number
of others. The size of the finite number depends on the context. For example,
suppose (G,O) is k-transitive not (k + I)-transitive, and r is a proper Jordan set.
Let /31, /32, ... ,/3k be k distinct elements in O\r with ex E r. Then define r 1 as
the maximum Jordan set containing ex and excluding /31, /32, ... , /3k' That way,
r 1 ;2 r, and every Jordan set in 0\ {/32, ... ,/3d larger than r 1 must contain (31. It
may well be that every Jordan set in 0\ {/32, ... , (3k} larger than r 1 must contain
other elements A besides /31, The collection of /31 together with all such elements
A (if any) now plays the role in the infinite case that the minimal increments in
finite Jordan groups play (compare with Remark (a) above). They are like "new"
minimal increments in the infinite case; and they point the way to invariant relations
in (G,O).

5 Infinite Bounded Permutation Groups and Comments on

Other Works

The starting point of the whole theory was the work of Jordan. One of his important
Theorems was the following Theorem of 1875 in [ 19].
Theorem 5.1 There is a function f such that whenever a finite primitive permu-
tation group (G,O) contains a permutation h E G with h of 1, Isupp(h)1 = m and
10 I > f( m), then G contains the alternating group (see Definition 2.1.5 for support
of a permutation).
Next comes the following generalization of Theorem 5.1 to infinite groups by
Wielandt [31].
Theorem 5.2 ff(G,O) is an infinite primitive permutation group and G contains
some permutation with finite support, then G contains the alternating group.
Comment on proof of Theorem 5.2.
Wielandt first produced a finite Jordan set in (G, 0), then used a variant of the
result in Lemma 4.1.5 to show that the subgroup of G consisting of maps with finite
support is highly transitive and thence concluded that G contains the alternating
group. How he produced the Jordan set went like this. Let Go be any finite
subgroup with finite support. Expand Go to a larger finite subgroup H of G such
that the support of each element in H is finite, and any block of H that contains
one point in supp( Go) contains supp( Go) as a subset. If E is the minimal block of
H containing supp(G o), then (HE,E) is primitive. (Here, HE is the action of H
on E). Moreover (Grr) := (h- 1 Gohlh E HE) is a normal subgroup of HE. Hence
it is transitive on E. Thus E is a finite Jordan set.
Definition 5.3. Let 0 be infinite. A permutation 9 of 0 is said to be finitary if
the support of 9 is finite.
Infinite Jordan Permutation Groups 167

Denote by SF(n) the group of all finitary permutations of n.

Theorem 5.4 [26, Theorem 3] Let (G,n) be an infinite group of finitary permu-
tations which is transitive but not primitive and which has a maximal congruence.
Then each class of the maximal congruence is finite, and

W' ::; G::; W

where W = H Wr SFp is the wreath product of Hand SFp, W' = commutator
group of W, p = number of congruence classes of the maximal congruence (= an
infinite cardinal), H corresponds to the action on a congruence class, and SFp is
the symmetric finitary group on the congruence classes permuting them as blocks.
One possible generalization of finitary permutations to higher cardinals is as
follows.
Definition 5.5. Let (G,n) be an infinite permutation group, and m be an infinite
cardinal number. Say G is m-bounded if ('rig E G)(lsupp(g)1 ::; m).
Assume the ZFC axioms of set theory. Let m+ be the successor cardinal of m,
and n:= 10,1.
The following problem naturally arises as an extension of Wielandt's Theorem
5.2.
Problem 5.6. Assuming n > m ::; ~o, classify all primitive m-bounded permu-
tation groups (G,n) withlnl = n.
Its solution contains more possibilities than the finitary case in Wielandt's
Theorem.
Theorem 5.7 (in [1]) Let (G,n) be an infinite primitive m-bounded permutation
group. Then (G, 0,) is a Jordan group containing primitive Jordan sets. Moreover,
it is highly transitive or leaves invariant one of these relations: a dense linear order,
a dense upper semilinear order or a dense C -relation. (A Sketch proof is in section
5.9 below).
5.8 Remarks on related work.
5.8.1. Droste's work on partial orders. Peter Neumann and Adeleke were quite
aware, at the early stages, of highly transitive examples of m-bounded groups built
on the product topology, and also of a simply primitive example (automorphism
Group of Q x wI). But the question lingered as to the existence or non-existence of
other examples that are not highly transitive. That was until December 1984 when
they constructed the example on the automorphism group of the semilinear order;
they were quite unaware of the work of Manfred Droste on partial orders [13] which
was published in 1985. In that work, Droste gave an impressive classification and
characterization of partially ordered structures using their automorphism groups as
tools. His groups certainly satisfy the properties of what are called Jordan groups
but he made no explicit mention of that fact.
168 S. A. Adeleke

5.8.2. Cameron's work on Combinatorics. Still on the surprising connections

to other works, we turn to the articles of Peter Cameron on combinatorics. In his
deep work on counting finite structures [8], Cameron constructed in 1983 an infinite
group (H, A) with the following properties:
(i) the number of orbits of H on k-element subsets in Sl equals the number of
rooted binary trees with k end-vertices,
(ii) (H, A) is 3-homogeneous, 2-transitive but not 2-primitive, and
(iii) (H, A) is universal in the sense that every 3-homogeneous and 2-transitive
but not 2-primitive group of countable degree is isomorphic to a subgroup of (H, A)
or to a subgroup of the group of order-preserving or -reversing permutations of Q.
He went further. He constructed a group (J, Sl), a one-point extension of (H, A)
which is 5-homogeneous, 3-transitive but not 3-primitive such that the number of
boron trees (all vertices have valency 1 or 3) with k end-vertices equals the number
of orbits of J on k-element subsets of Sl. He then asked specifically what the
connection of J to the infinite trivalent tree might be. Related to all these was the
work of Covington [12] (see also [10]) on N-free graphs. Cameron's question was
answered by Neumann and Adeleke in a joint seminar in 1985 and was reported by
Cameron in [10). The answer was that (J, Sl) was actually the automorphism group
of the set of points at infinity ("leaves") of a general betweenness relation induced
by a semilinear order. The names B, C, D relations came much later. Cameron's
work has connections with several areas including chemistry.
5.8.3. Application to Philosophy of Mathematics. The work on semi linear order
(1)-[3] had an application also to the philosophy of mathematics [5] answering one of
Frege's questions. The question pertained to the independence of Frege's axioms in
constructing real numbers. The automorphism group of a semilinear order provided
the answer.
5.8.4. Geometric Jordan Groups. Notice that the classification Theorem 3.1
in one case stopped at a Steiner system. A finer classification of this case would
be greatly valued. What conditions make the Steiner system an affine space or a
projective space need to be explored. For finite Jordan groups, Kantor [20] and
Neumann [27) showed that the Steiner system has to be affine or projective or a
space of one of the Mathieu groups or connected with the linear space one way or
another. But they used the classification of finite simple groups in pinning down
the detailed structure of the Steiner system.
For infinite Jordan groups with enough cofinite Jordan sets (that is, Jordan
sets with finite complements), Evans (16), without the classification of finite simple
groups, showed that the invariant relation is connected to a linear space or is highly
transitive. Zilber's work (32) went further, showing that in finite Jordan groups
with dimension six or greater, the relation has to be linked to a projective or an
affine space. That, of course, was without using the classification of finite simple
groups. Earlier, with the use of the classification of finite simple groups, Cherlin
and Zilber [11, Theorem 2.1], and Neumann [27, section 4) independently proved
Infinite Jordan Permutation Groups 169

the existence of a link between infinite Jordan groups having cofinite Jordan sets
and groups of the linear space.
5.8.5. Other related work. Other publications close to the contents sum-
marized in this paper include the work of McDonough [24] on the classification of
Jordan groups with perfect Jordan sets. Droste, Holland and Macpherson [15] gave
a description of the normal subgroups of the automorphism groups of semilinear or-
ders. Droste also did extensive work on normal subgroups of automorphism groups
of linear orders (see [14], for example). Wiegold [30], Giorgetta [17], and Segal [29]
worked on finitary permutation groups in a way connected to the contents of this
paper.
5.9 Sketch proof of Them·em 5.7
The following lemma connects bounded groups to Jordan groups.
LeITlITla5.9.1 An infinite transitive m-bounded group (G,O) contains a proper
Jordan set. If (G, 0) is primitive, then it contains a primitive Jordan set.
Sketch proof. We will assume that (G,O) is primitive since the reasoning for
the ordinarily transitive case is similar. Let a E O. Put ro := {a}, Ho {I}.
Expand recursively and in a natural way (Hi, r i ) i E N such that
(a) ro c r 1 c r 2 c ...
(b) Ho < H1 < H2 < ...
(c) every block of H i +1 that contains a contains r i, and
(d) ri+1 = r i U U{supp(g)lg E Hi+d·
For details of (c), pick any {3, /, 8 E rj\{a} with a oj:. {3, / oj:. 8. By primitivity of
(G,O), there exist sequences g1, g2, ... ,gk E G, and 7]0, 7]1,··.,7]k EO such that
7]0 = /, 7]k = 8 and {a,{3}gp = {rp-l,/p} for p = 1,2, ... ,k. For each such triple
(3, /,8, pick a corresponding sequence g1, g2, ... ,gk with the above properties and
denote by II({3, /, 8) the set of g1, g2, ... , gk picked. Define

Hi+l = Hi U U{II({3, /, 8) I {3, /, 8 E 0, {3 oj:. a, / oj:. 8}.

00 00

Now put I: := U ri and H = U Hi. Then H is primitive on I: and so I: is a

i=O ;=0
primitive Jordan set in 0 with II:I ::; m. Since II:I < m+, the successor cardinal to
m, which is less than or equal to n, then I: is a proper primitive Jordan set.
5.9.2. Simply primitive case of Theorem 5.7.
Assume (G, 0) is simply primitive and m-bounded.
By Lemma 5.9.1,0 contains a proper primitive Jordan set r. Let a E rand
{3 E O\r. Define A := M PJ({3; a)~that is, the maximal primitive Jordan set in 0
containing a and excluding {3. Its existence follows from Lemma 4.1.2. If

(5.9.3) (3J.l, v E O)(J.l oj:. v)('r/g E G)(v E Ag -+ J.l E Ag),

170 S. A. Adeleke

then define v > II for such pairs. It is easy to see that :S is G-invariant, reflexive
and transitive. Primitivity of (G, rl) implies that :S is antisymmetric and also that
:S satisfies condition 2.2.2(i) of an upper semilinear order in section 2. The use of
the stabiliser G(n\Ax)-aS defined in 2.1.9-shows that :S satisfies condition 2.2.2(ii)
as well. Denseness of < holds because no Jordan set can have a maximal element.
Thus (G, rl) preserves a dense upper semilinear order (which is possibly a linear
order) on rl if (5.9.3) holds.
Suppose condition (5.9.3) fails but the following holds:

(5.9.1) (\lgEO)(Ag2A or Ag~A or AgnA=¢).

Then define for any 0:,;3,) in rl,

C(ct;;3,) if and only if (:3h E 0)([3,) E Ah tw f/:. Ah).

It follows easily that the defined relation C above satisfies conditions (C1)-(C4) of
a C-relation in (2.2.5). Axiom (C5) holds since otherwise the relation ~ defined by
WI ~ W2 if and only if

is a proper congruence of (0, rl). And denseness follows because A and its translates
under G are primitive.
Lastly, we claim t.hat if (5.9.3) fails, then (ii.9.1) holds. We shall supply some
deta.ils here since the method of proof in [1) is different. but. more general. Suppose,
with hope of a contra.diction, that (.5.9.3) and (5.9.4) both fail to hold. Then

(5.9.5) (:3g E G)(Ag nA:f ¢,Ag\A:f ¢ and A\Ag:f ¢).

In other words, A and Ag are typical. It will be considerably helpful to the reader
to draw a diagram here. By definition of A, Ag = Af Pl(;3g; o:g). Since Ag U A is
then a primitive Jordan set larger than A and Ag, then [3 E Ag\A and [3g E A\Ag.
Furthermore, (;3g)g E Ag since ;3g E A. Pick h E O(n\Ag) such that ;3g2h = ;3.
Since ;3g f/:. Ag, t.hen ;3gh = [3g, and so ;3(gh)2 = ;3ghgh = ;3ggh = ;3g2h = ;3.
Moreover Agh = Ag by definition of h. Thus, because we can change 9 to gh, we
can assume without loss of generality that ;3g2 =;3. Then since A = M P 1 (;3; ;3g)
and /3g 2 =;3, it follows that Ag2 = MP1(;3g2;;3g3) = MP1(;3;;3g) = A.
Define
~(A) := {~ E rllA = M Pl(~; o:)}.

That is, A is maximal with respect to excluding every element in ~(A). Put another
way, every primitive Jordan set properly containing A must contain ~(A). (Note
that the definition of ~(A) and the arguments in this paragraph can be extended
to Jordan groups beyond bounded ones. We put it here because of our view that
the reader who understands the concept in t.his simpler setting can extend it easily
 Infinite Jordan Permutation Groups 171

to other cases.) Of course {3 E il. Then, in a way, the set il corresponds to the
minimal increment described in Remark 4.1.6(a) on finite Jordan groups. Just as in
the finite case, it is easy to show that Ll(A) is a block for G(n\E) where ~ := AUAg.
But, by Lemma 4.1.3, ~ is a primitive Jordan set. Thus Ll(A) must be precisely
{{3}.
This fact about il(A) implies in the following way that ~ = 0 and (G,O) is
2-transitive.
Let J.l E ~\{{3}. If J.l E A, then MPJ({3;J.l) = A with ILl(MPJ({3;J.l)1 = 1.
Suppose J.l E Ag\(AU{{3}). Since il(A) = {{3}, J.l rt. il(A). So, if we pick Z E G(n\Ag)
such that ·{3z = /-" then Az ;2 AU {;3} and Az C ~. Pick x E G(n\Az) such that
{3gx = {3. Then /-'X = /-" Agx C ~ and Agx = M PJ({3; /-'). Then from lil(A)1 = 1,
we deduce lil(Agx)1 = 1 and so ILl(M PJ({3; /-'))1 = 1. Thus for any J.l E ~\{{3},
ILl(M PJ({3; /-')1 = 1.
Suppose now that 0 f:. ~ and that, unlike /-" some element v exists in
O\~. Since we are assuming that (5.9.3) fails for A, then M PJ({3; v) exists. Let
r := M P J({3; v). From v rt. A, we deduce that r and A are disjoint both being
maximal primitive Jordan sets with respect to excluding {3 but with v E f\A. Fur-
thermore, r n Ag = cp otherwise r U Ag will be a primitive Jordan set larger than
Ag and excluding {3g and then Ag will cease to be M P J({3g; {3). Thus, r n ~ = cp.
By use of maps in G(n\E), we see that il(r);2 L That is, lil(MPJ({3;v))1 > 1 : a
result different from that of Ll( M P J ({3; J.l)) for any J.l E ~\ {{3}. We conclude then
that v rt. /-,Gf3 for any /-' E ~\{{3}. Thus, ~Gf3 = ~. And for any w E G(n\Ag)
with ;3w f:. ;3, we see that ~w = ~. So, ~(Gf3' w) =~. But then, G = (Gf3, w)
by primitivity of G. Hence ~G = ~ and ~ = 0- contradicting our supposition of
existence of v. We deduce therefore that ~ = O.
e
From ~ = 0, we now deduce that (G,O) is 2-transitive. Let be any element
in A\(Ag U {{3g}). Then eg E Ag\(Ag 2 U {{3g2}) = Ag\(A U {{3}) since Ag2 =A
and {3g2 =;3. Exactly as done for /-' when assumed to be in Ag\(A U {{3}), we
can construct x' E G such that (eg)x' = eg, ({3g)x' = {3. Then egx' = eg and
({3g)gx,-l = {3x,-l = {3g. Thus gx,-l E G(f3g) and e(gx'-l) E Ag. Therefore,
eGCf3g) nAg f:. cp. Hence e lies in the orbit of G(f3g) containing Ag. Since e
is
arbitrary in A\(Ag U {{3g}, then (G, 0) (which is the same as (G, ~)) is 2-transitive.
This contradicts our hypothesis. Thus (5.9.4) must hold if (5.9.3) fails, and so
Theorem 5.7 holds when (G, 0) is simply primitive.

5.9.6. The 2-transitive not 3-transitive case. If (G, 0) is 2-transitive but not
3-transitive, then G has an invariant C-relation. Its proof is similar but simpler
than the proof of the simply primitive case. The proof defines A as the maximal
2-transitive Jordan set containing a and excluding {3 and shows that (5.9.4) always
holds as follows. If (5.9.4) fails, define ~ = AUAg as before. Then ~ is 2-transitive.
Pick /-' E A\Ag and v E Ag\A. Then by use of GC n \A)(Il) followed by use of
G Cn \Ag)(lI), we see that G Cn \E)CIl)(lI) is transitive on ~\{J.l, v}. Thus ~ is 3-transitive
which gives a contradiction.
172 S. A. Adeleke

5.9.7. (G,O) t-transitive for t ~ 3. Consider a maximal proper t-transitive

Jordan set A. Since G is t-transitive with t ~ 3, then if A1, A2 E A, A1 f:. A2, and
A3 E O\A, there is g E G such that )'1 g = A1, A2g = A3 and A3g = A2. Thus, the
situation in (5.9.5) always occurs. This will imply that AU Ag is a (t + 1)-transitive
Jordan set just as in the last case. (In the last part of the argument, pick 11 E A\Ag,
1.1 E Ag\A as before and pick any other t - 2 elements in A which leave out at least
one element in An Ag). Thus G is (t + 1)-transitive by Lemma 4.1.4. By induction,
(G,O) is highly transitive.
5.1 0 Transitive not primitive m-bounded groups
Lelllllla 5.10.1 Suppose (G,O) is a transitive m-bounded group. If P is a proper
congruence, then Ipi ::; m.
Reason: If g E G and 1jJ is a block with 11jJ1 > m, then Isupp(g)1 ::; m and so 1jJ\
(supp(g)) f:. ¢. Therefore 1jJ n 1j!g f:. ¢; so 1jJg = 1jJVg E G. Therefore 1jJ = O.
The following are also not difficult to prove.
Lelllllla 5.10.2 [1, Lemma 2.2] Suppose that G is transitive and m-bounded and
that there is no maximal proper congruence on O. Then
(z) n = m+;
(ii) there is a properly increasing chain Po < PI < ... < P~ < ... of proper con-
gruences indexed by 1.1, the initial ordinal of n, and UE<vPE is the universal
relation,-
(iii) if 6. ~ 0 and 16.1 ::; m then there exists 11 < 1.1 such that all members of 6. are
equivalent modulo PI';
(iv) if P is any proper congruence then P ::; Pry for some 1] < 1.1.
Definition 5.10.3 Let (G, 0) be a transitive m-bounded permutation group. Say
(G, 0) is almost primitive if (G, 0) has a maximal proper congruence pmax with each
class of the congruence finite.
Lelllllla 5.10.4 [1, Lemma 2.3] IfG is a transitive, almost primitive, m-bounded
group of permutations of 0 then the maximal proper congruence pmax is unique.
Remark 5.10.5. By Lemma 5.9.1, a transitive m-bounded group is a Jordan
group.

6 Infinite Primitive Jordan Groups with Proper Primitive

Jordan Sets

Bounded groups discussed in Theorem 5.7 constitute a type of infinite primitive

Jordan groups with proper primitive Jordan sets as in Theorem 5.7. A natural
problem is then to classify all primitive Jordan groups that contain proper primitive
Jordan sets. This question is more complex than that of m-bounded groups. For
Infinite Jordan Permutation Groups 173

we can no longer establish existence of proper t-transitive Jordan sets in t-transitive

Jordan groups for t ~ 2 as is done in the proof of Lemma 5.9.1. But the final result
is this.
Theorem 6.1 (see [2]) Let (G, 0.) be an infinite primitive Jordan group that con-
tains a proper primitive Jordan set. Then (G, 0.) either is highly transitive or leaves
invariant one of the following relations on 0. :
(a) a dense linear order, a dense semilinear order or a dense C -relation if (0,0.)
is simply primitive;
(b) a dense linear betweenness relation, a dense general betweenness relation, or a
dense C -relation if (0, n) is 2-transitive not 2-primitive;
(c) a dense circular order or a dense D-relation if (0,0.) is 2-primitive not 3-
transitive;
(d) a dense separation relation or a dense D-relation if (0,0.) is 3-transitive not
3-primitive.
In the case when (G,0.) is 3-primitive then it is highly transitive.
Remark 6.2. The proof of Theorem 6.1 in [2] uses Theorem 6.3 below by
Cameron. We will roughly indicate below an alternative proof of Theorem 6.1 not
using Theorem 6.3. The aim is to present the reader with an elementary proof
which when combined with the version in [2] produces good understanding.
Theorem 6.3 (Cameron) [8]) Let (0,0.) be an infinite highly homogeneous group
which is t-transitive not (t + I)-transitive. Then t S; 3 and 0 is a group of order-
preserving or reversing transformations of a linear or circular order on 0..
6.4 Sketch proof of Theorem 6.1
6.4.1. (0, n) simply primitive. This is proved by the argument.s for simply primitive
m-bounded groups in (5.9.2). Before we pass on to the next case, let us make one
observation.
Lelllllla 6.4.2 Let (0,0.) be an infinite simply primitive permutation group with
a proper primitive Jordan set. Suppose 0. contains a maximal primitive Jordan set
A which does not induce a O-invariant linear order through property (5.9.3). Then
no two proper primitive Jordan sets r l , r 2 , are typical. That is, the following is not
possible:

Sketch proof of Lemma 6·4.2. Let A := M P J(j3; a) as in the proof of Lemma

5.9.2. Assume that A induces an invariant nonlinear upper semilinear order on n
through property (5.9.3). It is not hard to see that nonlinearity of the semi linear
order implies that if r is a Jordan set in n containing an element ~, then r also
contains an element 1] incomparable to ~. So, the map in 0C!1\r) that moves ~ to
1] also moves all elements below ~. Hence r contains all elements below ~. Thus
174 S. A. Adeleke

every Jordan set is a branch below a branch point (as defined in (2.2.2» in the
semilinear order or a union of branches below branch points. And every primitive
Jordan set in fl is a branch or a union of an increasing chain of branches. Since no
two branches are typical, we conclude then that no two primitive Jordan sets are
typical. Hence the lemma holds if G preserves a nonlinear semilinear order on fl.
A similar argument proves the lemma when G preserves a C-relation on fl.
6.4.3. (G, fl) 2-transitive not 2-primitive. If (e, fl) has a proper 2-transitive
e
Jordan set, then the arguments in section 5.9.6 will suffice to show that preserves
a C-relation on fl.
Let us assume then that (e, fl) has no proper 2- transitive Jordan sets. Then
the proof for this case is a small step beyond that of simply primitive m-bounded
groups in section 5.9.2. Let A:= MPJ«(3;a). Since (e,fl) is not 2-primitive, then
A "# fl\ {(3}. If condition (5.9.4) holds, thene preserves a dense C-relation on fl.
If condition (5.9.4) fails, then as before, AU Ag = fl for some gEe.
In that case, for every a, (3" E fl, define B(a; (3,,) if and only if

(01=,) or (01=(3) or (MPJ(a;(3)nMPJ(a;,) = 0).

It follows easily that the defined relation B satisfies axioms (Bl), (B3), and
(B5) of a general betweenness relation in 2.2.3. As for axiom (B2), suppose
B(a;(3,,) with 01,(3 and, distinct. Then (3 f/: MPJ(o:;,). Because the defined
set ~(MPJ(a;,)) := wMPJ(e;,) = MPJ(o:;,)} equals the singleton set {a}
(in a way similar to the procedure in section 5.9.2), then from (3 f/: M P J( a; ,), we
deduce that M P J«(3;,) is larger than M PJ(a; ,). Thus 0: E M PJ«(3; ,). Therefore
....,B«(3; a, ,). So axiom (B2) holds as well. We can deduce more. The arguments
show that every maximal primitive Jordan set is contained in a larger maximal
primitive Jordan set. Equivalently by 2-transitivity, every maximal primitive Jor-
dan set contains a smaller maximal primitive Jordan set. So, to prove (B4), sup-
pose ....,B(o:;(3,,). Then IE MPJ(a;(3). Put r:= MPJ(a;(3). Then r contains
a smaller maximal primitive Jordan set ~ where ~ is a maximal primitive Jor-
dan set M P J (J-l; l/) in fl containing some element l/ and excluding some J-l. Hence
(GCrl\r), r) is a simply primitive Jordan group, simply primitive because we have
ruled out the case of proper 2-transitive Jordan sets. Whichever of the types of
invariant relation in (5.9.2) that ~ induces on r, it is easy to see from the defini-
tions of the relations that there exists a translate ~x C r, x E GCrl\r) such that
1', (3 E ~x. Thus r contains properly some M P J (8; (3) with, E M P J (8; (3) and
6 E r. In that case, B(6; 0:, (3) and B(6; 0:,,) and so (B4) holds. Consequently, G
preserves a general betweenness relation B which may reduce to the special case of
a linear betweenness relation.
6.4.4. (G, fl) 2-primitive not 3-transitive. Let 8 E fl. Then (G 6 , fl\{6}) is
simply primitive.
We first show that on fl\ {8}, no maximal primitive Jordan set in fl\ {8} induces
a nonlinear upper semilinear order that is invariant under G 6. Dugald Macpherson
 Infinite Jordan Permutation Groups 175

[23] in a preprint also proved the nonexistence of any Go-invariant nonlinear semi-
linear order on O\{8} if (G,O) is 2-primitive but his proof is not the same as the
one below. He showed further that there is no one-point extension of a primitive
Jordan group (H, r) if H preserves a general betweenness relation on r, or preserves
a D-rclation on r. But again his proofs are not the same as those in this paper.
Suppose for a contradiction that there is a maximal primitive Jordan set in
0\ {8} which induces a nonlinear upper semilinear order on 0\ {8} invariant under
Go. Define for every W1,W2 E O\{8}, WI '" W2 if and only if there is a primitive
Jordan set r in 0 containing 8 and excluding WI and W2. Clearly, '"" is reflexive,
symmetric and Go-invariant. To check if '" is transitive, suppose WI '"" W2 and
W2 '" W3. Then there exist primitive Jordan sets r 1 and f2 in 0 such that 8 E r 1 nr 2 ,
W2 f{. r 1 U r 2, WI f{. r2 and W3 f{. r 2. In other words, 1'1 and r 2 are non-disjoint
primitive Jordan sets in O\{W2} with its nonlinear semilinear order. By Lemma
6.4.2, either f1 ~ r 2 or r 2 ~ r 1 . In either case, the smaller one contains 8 and
excludes WI and W3. Thus, WI '"" W3 and,"" is transitive. So, '" is a congruence
which must then be universal by primitivity of (Go, 0\ {8}). But that means that if
W E 0\ {8}, then for any w. E 0\ {8} with w. < 8 in the nonlinear semilinear order
on O\{w} there is a primitive Jordan set r containing 8 and excluding w. and w.
Since r is Jordan, with 8 > W., 8 E 1', w. rt. r, then each element of r is an upper
bound of w •. In that case, r contains some A such that 8 > A > W •. Let I; be the
primitive Jordan set guaranteed by < on 0\ {w} which contains A and omits 8 and
w. Then w. E I;\r, A E I; n r, and 8 E r\I;. Then 1; and r will form a typical
pair both excluding w. This contradicts Lemma 6.4.2. Hence Go does not preserve
a nonlinear semilinear order on 0\ {8}.
Suppose Go preserves a C-relation on O\{8}. With the same method as
in Section 5.9.2 (now applied to (Go, O\{8}), let A := MPJ(8,(3;o} Also let
~1,6 E O\(A U {(3}) with ~1 "# 6· The method of the sketch proof of Lemma
6.4.2 ~hows that a union of an increasing chain of translates of A equals O\{6},
and a union of another chain of translates of A equals 0\ {6}. It then follows that
there are two translates of A, one picked from each chain, which are typical. As a
consequence, A and Ah are typical for some h E G. By Lemma 6.4.2, Au Ah = O.
So in case of a C-relation on 0\ {8}, 0 contains typical pairs of translates of A with
the union of any such pair being O. This property on typical pairs prompts the
following definition. For G', (3",8 E 0, define

D(G',(3;" 8) if and only if

(3x E G)((G', (3 E Ax) /I. (-y, 8 f{. Ax)) or (3y E G)((G', (3 f{. Ay) /I. C" 8 E Ay)).

Then, relation D above satisfies all the axioms of a D-relation contained in 2.2.6
with a little use of Lemma 6.4.2. It is a dense D-relation because of the primitivity
of A.
Lastly, suppose Go preserves a linear order < on 0\ {8} through property
176 S. A. Adeleke

(5.9.3). In that case, define for every distinct " ~, TJ in 0,

p(J,~, TJ) if and only if

(J =~) or (~= TJ) or (TJ = ,) or (Vx E G)(TJX = 8) ~ (,x < ~x)).

Then, it is clear that p is G-invariant and satisfies axioms (p2) and (p4) in 2.2.7.
As for axiom (pI), suppose in hope of contradiction that p(a, (3, ,) and -,p«(3", a).
Then a, (3" are distinct and there exist x, y E G such that

(,x = 8) /\ (ay = 8) /\ p(ax, (3x, 8) /\ -,p«(3y, ,y, 8).

By linearity of <, we can rewrite condition on p as p(ax,(3x,8) and p(Jy,(3y,8).
From condition (5.9.3), we see that the primitive Jordan set A is an initial segment
of the linearly ordered set (O\{8}, ~). By usual arguments (see first paragraph of
proof of Theorem 9.3.2(i) below), Gh is order 2-transitive on O\{8}. Thus, there is
z E Gh such that axz = ,y and (3xz = (3y. Then (3(xzy-l) = (3, a(xzy-l) =, and
,(xzy-l) = 8zy-l = 8y-l = a. Therefore, the map xzy-l fixes (3 but interchanges
, and a. Thus it does not preserve the linear order on 0\ {(3}, a contradiction. So,
p satisfies condition (pI) also. Now, to condition (p3);
p(a, (3, ,) /\ -,p(8, (3, ,) -- p(a, (3, ,) /\ p«(3, 8, ,)
using the definition of p, then the linearity of <, and then the G-invariance of p;

using (pI)
~ p(8, a, (3)
using the definition of p, followed by the definition of < on 0\ {8} through 5.9.3,
and then the G-invariance of p.
So, (p3) holds. We conclude therefore that p is a G-invariant circular order.
6.4.5. (G,O) 3-transitive not 3-primitive. If 8 E 0, then G h is 2-transitive not
2-primitive on O\{8}. From 6.4.3, G a preserves on O\{8} a betweenness relation,
or a C-relation.
Suppose it is a C-relation. By 3-transitivity of G, A and Ag are typical for
some g E G. Then, arguments similar to those on the C-relation in 6.4.4, with
Lemma 6.4.2 suitably used, imply the existence of a G-invariant D-relation on O.
Suppose the invariant relation on 0\ {8} is not a C-relation but a nonlinear
betweenness relation. We shall show that this cannot occur. Let (31, (32 be distinct
elements of 0\ {8}. Then, of course, an invariant nonlinear betweenness relation
occurs on O\{(3d too. Since (G/31> 0\ {(3d) is not 2-primitive, and is 2-transitive,
then there is some maximal primitive Jordan set r := M PJ«(31, (32;'x) with respect
to excluding (31 and (32 which is disjoint from M P J«(31, (32; 8). So, r is a subset of
Infinite Jordan Permutation Groups 177

0\ {o}. (The reader will benefit from two diagrams here; one for the betweenness
relation on 0\ {o}, another for the betweenness relation on 0\ {,8d.) Whatever the
location of r in the betweenness relation on 0\ {o}, one thing is clear. There is a
primitive Jordan set II C 0\ {o} disjoint from r which contains one of ,81,,82 and
excludes the other. Without loss of generality, assume ,81 E II and ,82 ~ II. By use
of GCO\IT), we deduce that r = M P J(w, ,82; A) for all w E II. But this contradicts
the defined set t::..(M P J(;31,;32; A)) := WM P J(';, ;32; A) = M P J(;31, ;32; A)} being a
singleton set as described in 6.4.3. Thus Go cannot preserve a nonlinear betweenness
relation on O\{O}.
The only other possibility on 0\ {o} is a linear betweenness relation invariant
under Go. This leads to a G-invariant separation relation on 0 proved in [2) using
Cameron's Theorem. But there is an alternative proof not using Cameron's Theo-
rem. It is messy though. The proof, like others so far, relies on earlier arguments;
in this case, on the arguments for the 2-transitive not 2-primitive possibility. Let
0, w, 0: be distinct in O. Let A := M P J( 0, w; 0:). Since we are assuming that no G o-
invariant C-relation exists on 0\ {o}, then from the first two paragraphs of section
6.4.3 now applied to (Go), 0\ {5}), we note that AUAg = 0\ {5} for some g E GJ. It
follows that A is not a bounded interval with respect to any of the two dense linear
orders associated with the linear betweenness relation on 0\ {5} assumed invariant
under GJ. So, for one of the two linear orders, < (say), A is an initial segment.
Define HI := {g E Go I Ag ~ A or Ag ~ A}. Then clearly, (H1,0\{5}) becomes a
simply primitive group. Moreover < satisfies:

(6.4.6) p < v if and only if (Vh E HI)((v E Ah) --4 (p E Ah)).

Next, observe in the following arguments, that M P J (w, 0:; 5) consists of 5, the el-
ements less than 0: in the linear order <, and the elements greater than w. For
if ~ is any element in 0\{5} less than 0:, then M PJ(5, 0:;0 is the set of all el-
ements lower than 0:. The result on t::.. being singleton in 6.4.3 now shows that
MPJ(5,0:;~) =F MPJ(w,o:;O but rather MPJ(5,0:;';) C MPJ(w,o:;O. Thus
OEM P J(w, 0:; ~). Similarly, if 1] > w, then 0 E M P J(w, 0:; 1]). So, indeed,

MPJ(w,0:;5)={-yEOIr<0: or i>W}U{O}.

This of course extends to MPJ (;31, ;32;8) for any two distinct ,81,,82 in 0\ {5}. All
these show that if we define

then we can through lengthy but very elementary steps show that (II, 0) is a 2-
primitive group preserving the circular order associated with < . With that proven,
if we pick x E G O\H 1 , then x reverses < while keeping 0 fixed. In fact, x reverses
the circular order associated with < . Moreover, Go = (III, x) and consequently
178 S. A. Adeleke

G = (HI, H 2 , x) = (H, x). Therefore G preserves the circular order or reverses it.
That is, G preserves a separation relation on 0 if G6 preserves a linear betweenness
relation on 0\ {o}.
6.4.7. (G,O) 3-primitive not 4-transitive. This case really does not occur. The
arguments should, by now, be familiar. The group G 6 is 2-primitive not 3-transitive
on 0\ {6}. So, employ the notation in 6.4.4.
Suppose G6 preserves a D-relation on 0\ {6}. Then, for any 6,6, 6, ~4 distinct
in 0, MPJ(6,6,6;~4) exists since the invariant relation on O\{~1,6} is not a
semilinear order but a C-relation. Let A := M P J(6, f3,~; 0:). Every primitive
Jordan set in 0\ { 6, f3} is either a G 6/3- translate of A (that is, Ax for some x E G 6/3),
or is a union of an increasing chain of such translates (using the method of proof of
Lemma 6.4.2). Let {Aili E I} be an increasing chain of primitive Jordan sets, each
containing 0:, whose union is 0\ {o, f3}. Without loss of generality, assume that each
i in I has a successor i+. Let Ai E Ai+ \Ai. From the definition of the D-relation in
6.4.4, ifi < j in I and D 1 , D2 are the D-relations on O\{6} and O\{f3} respectively,
then

(6.4.8)

(A diagram will be useful here).

Let io E I and define

1'1:= MPJ(0:,6,Ai o ;f3), 1'2:= MPJ(0:,f3,Ai o ;6).

From (6.4.8), Ai rt. 1'1 U 1'2 if i < io. Let p, E 1'1 \{f3}. Since u{Aili E I} = O\{6, f3},
then

In that case, Aj E r 1, Vj > jo. Similarly, we can deduce that for some ko E I,
Ak E r 2 , Vk> ko. Hence Ap E r l n 1'2 for all p > max{jo,k o}. Thus r 1 ,r 2 are
typical primitive Jordan sets in O\{o:, Aio}' This contradicts Lemma 6.4.2. So, Go
cannot preserve a D-relation on O\{6}.
If on the other hand Go preserves a circular order on 0\ {6}, then G O/3 preserves
a linear order < on 0\ {6, f3}. Let 0: < '" in 0\ {6, f3}. Then the arguments in section
6.4.5 on the circular order show that

III:=MPJ(6,0:,,,,;f3)={f3}u{~I~<0: or ~>"'}
II 2 := MPJ(f3,0:,T);6) = {o} U {~I ~ < 0: or ~ > "'}
Thus III, II2 is a typical pair with III \II2 = {f3}. So, III U II2 is a proper 2-transitive
Jordan set excluding 0: and ",. Then (G 6/3, 0\ {6, f3}) becomes 2-transitive by Lemma
4.1.4. This contradiction shows that G 6 cannot preserve a circular order on 0\ {6}
either.
Infinite Jordan Permutation Groups 179

6.4.9. (G, 0) 4-transitive not 4-primitive. This case also is not possible.
Suppose (G, 0) contains proper 2-transitive Jordan sets. Then, as indicated in
the first statement in Section 6.4.3, the arguments in Section 5.9.6 are enough to
show that Go" preserves a C-relation on 0\ {6, J.t} for any distinct 6 and J.t in O.
Then the second paragraph in Section 6.4.5 shows that Go preserves aD-relation
on O\{6}.
In that case, replace maximal primitive Jordan sets in the arguments in section
6.4.7 by maximal2-transitive Jordan sets. Then f1 Uf 2 will be a proper 3-transitive
Jordan set at the end of the arguments making GC/,Ai o 3-transitive and thus (G, 0)
5-transitive, a contradiction.
Suppose then that (G,O) does not contain 2-transitive Jordan sets. If Go
preserves a D-relation on 0\ {6}, use arguments in 6.4.7 to get a contradiction. If,
however, Go preserves a separation relation on O\{6}, use the part of the arguments
in section 6.4.7 on circular orders to get a contradiction.
6.4.10. Conclusion to proof of Theorem 6.1. From sections 6.4.7 and 6.4.9, we
see that if (G,O) is 3-primitive then it is 4-primitive. When this result is in turn
applied to (Go, 0\ {o}), the latter is shown to be 4-primitive, and so (G, 0) is proved
5-primitive. Such an inductive argument shows that if (G, 0) is 3-primitive, then it
is n-primitive for all n, and so is highly transitive.

7 General Infinite Primitive Jordan Groups

(No assumption of presence of proper primitive Jordan sets.)

7.1. Sketch proof of Theorem 3.1 for Simply Primitive Jordan Groups
Define A := M J«(3; a), the maximal Jordan set in 0 containing a and exclud-
ing (3. If A satisfies (5.9.3) or (5.9.4), then the arguments in section (5.9.2) show
that A induces an invariant dense semilinear order or an invariant C-relation, not
necessarily dense.
Suppose, however, that A violates (5.9.3) and (5.9.4). If I.Q.(A)I = 1, then the
arguments of section (5.9.2) again yield a contradiction. Hence if A violates (5.9.3)
and (5.9.4) then I.Q.(A)I > 1.
From the property that I.Q.(A)I f:. 1, we now show that .Q.(A) and .Q.(A)x cannot
be typical for any x E G. Suppose (3 E .Q.(A) n .Q.(A)x and r E .Q.(A)x\.Q.(A). Since
.Q.(A)x = .Q.(Ax), then r E .Q.(Ax)\.Q.(A). So .Q.(Ax) f:. .Q.(A) and therefore A
and Ax are different maximal Jordan sets with respect to excluding the same (3.
Consequently Ax n A = ¢. If r ¢. A, then since r ¢. .Q.(A), we have M J(r; a) :J A,
and so .Q.(A) C M J(r; a); in particular (3 E M J(r; a). Put II := M J(r; a). Then
II and Ax are both maximal Jordan sets with respect to excluding the same r,
one containing A and the other disjoint from A. So Ax n II = ¢. Since (3 E II
and (3 E .Q.(Ax), then the use of GC!1\I1) yields .Q.(Ax) 2 II. But .Q.(A) C II.
Hence .Q.(Ax) :J .Q.(A). Suppose on the other hand that rEA. Consider the
180 S. A. Adeleke

typical pair A and Ag with (3 E Ag\A and (3g E A\Ag. Pick z E Gcnv\) such that
(3gz :::: i. Then Agz :::: M J(-y; (3) with Agz and A being typical. Now Agz and Ax
are both maximal Jordan sets with respect to excluding 'Y with (3 E Agz and (3 rf. Ax.
Therefore, Agz n Ax :::: ,p. This with the earlier deduction of A n Ax :::: ,p shows
that Ax n (A U Agz) :::: ,p. Since (3 E ~(Ax), then by using elements in G which fix
pointwise O\(A U Agz), we see that ~(Ax) ::2 Au Agz. But ~(A) c:;;; (Agz\A) since
(Agz U A) :J A. Hence ~(A) C (A U Agz) c:;;; ~(Ax). Therefore, ~(A) c:;;; ~(Ax).
lIenee, in any case, ~(Ax) :J ~(A) if ~(Ax) n ~(A) # ¢ and ~(Ax)\~(A) # ,p.
Thus ~(A) satisfies a property similar to property (5.9.4) on A as follows:

(7.1.1) (Vg E G)(~(A)g n ~(A) :::: ,p) V (~(A)g c:;;; ~(A)) V (~(A)g ::2 ~(A)).

And because ~(A) C Ag and A violates (5.9.3), then ~(A) also violates a property
similar to (5.9.3). Therefore if A violates (5.9.3) and (5.9.4), then ~(A) induces a
C-relation (not necessarily dense) on 0 which is invariant under G.
7.2. On general Theorem 3.1 when (G, 0) is 2-transitive
This paper gives no sketch proof of the general Theorem 3.1 for 2-transitive
.Jordan groups mainly because a very short section (section 4 of [4]) does so already.
The sketch proof in [4] focuses on the situation when no primitive .Jordan sets exist
in t.he Jordan group. It should pose no difficulty to a reader who has gone over the
sketch proofs in this paper. Even the full proofs in [4] themselves should not be
difficult since they are natural extensions of the reasoning described in the earlier
sections above. Unexpected examples occur in the general case, of course; see for
instance [6].

8 On Invariant Relations in Primitive Permutation Groups

8.1. Introduction
From an inverted tree, three other relations can be defined besides the order on the
points on the tree (see[2]' [3], [4] or [6]. Firstly, a betweenness relation exists between
the points on the tree. Secondly, apart from the directly upward and directly
downward directions on t.he tree, there are other directions. An abstraction of the
properties of the directions is called a f)-relation. Thirdly, the relation induced on
the maximal chains on the tree is called a C -relation. Precise definitions of these
relations are given in section 2.2 above.
This section addresses the conjecture in [3] on when a simply primitive infinite
permutation group possesses invariant relations connected with an inverted tree.
Specifically, the work here states that the conjecture is true if "betweenness relation"
is relaxed to "B-relation". A B-relation is like a betweenness relation on an inverted
tree except that two incomparable clements in the tree may have none of their
common upper bounds located between the two elements.
Infinite Jordan Permutation Groups 181

Definition 8.2. A ternary relation satisfying (B1), (B2) and (B3) in (2.2.3) above
is called a B-relation.
Definition 8.3 Let (G,O) be a permutation group and let E be a subset of 0
satisfying lEI ~ 2, 10\E/ ~ 2. Then E is said to be syzygetic with respect to G if

('ttg E G)«(E nEg) = <fJ) V (E U Eg = 0) V (1: ~ Eg) V (Eg ~ E)).

It is said to be hypersyzygetic with respect to G if

('ttg E G)«(E nEg) = ¢) V (E ~ Eg) V (Eg ~ E».

Conjecture 8.4 The conjecture [3, conjecture 34.8] states the following:
Let G be a primitive permutation group on o. If there exists a subset of 0
that is syzygetic with respect to G, then 0 carries a G-invariant semilinear order
relation, betweenness relation, C -relation or D-relation.
Observation 8.5. Note that if E is a subset of 0 which is syzygetic with respect
G, then O\E is also syzygetic with respect to G by DeMorgan's laws of set theory.
Observation 8.6. If a and {J are distinct elements of 0 and if there is some
translate1:g of E containing {J and excluding a, then by the syzygetic property of E,
the collection of all translates of E containing {J and excluding a is a chain. We define
the union of the chain as ME(a; {J) where 'M' is to connote some 'maximality'.
More precisely, we define

ME(a; {J) := U{Eg I 9 E G, {J E Eg, a f!. Eg.}

We repeat that M1:( a; {J) fails to exist if no translate of 1: contains {J and excludes
CI'. Similarly, if a, {J, and, are distinct, we define

ME(CI', {J; ,) := U{Eg I 9 E G" E Eg, a, {J f!. Eg}

The set M1:(CI' , {J; ,) may also fail to exist.

8.7 Results and Comments
Lemma 8.7.1 Let (G, 0) be a permutation group and let 1: be a subset of 0 which
is syzygetic with respect to G. Assume the following:
(Hi) (VCI',(3 E O)«CI',(3, distinct) ---+ (3ME(CI'; (3»(M1:(CI'; (3) "# O\{CI'}»
(H2) (3C1', {J" E O)(CI', (3", distinct)«3ME(CI', {J;,» 1\ ClIME(CI', ,; (3».
Then there is a G-invariant B-relation on O.
Remark 8.7.2. The conclusion in the last lemma cannot be strengthened to a
betweenness relation (see section (8.8) below).
182 S. A. Adeleke

Lemma 8.7.3 If (H1) in Lemma 8.7.1 holds and (H2) fails to hold, then

(S.7.3.1) ('10:, {3" E r.!)(o:, {3, ,dIstinct )(:3g E C)((o:, {3 (j: 'L.g) 1\ (T E 'L.g»

Remark 8.7.4. Assume (C, r.!) is primitive and 'L. is a syzygetic subset of r.! with
neither 'L. nor r.!\'L. hypersyzygetic with respect to C. If condition (S. 7 .3.1) above
holds, then G preserves a D-relation on r.!.
Lemma 8.7.5 Lemma 8.7.1 above shows that Conjecture 34.8 in [3] (stated in 8.4
above) holds if "betweenness relation" is changed to "B -r'elation" in the conjecture.
S.S Proof of Results in section 8.7
Proof of Lemma 8.7.1. Assume the hypothesis of the Lemma. We first show that a
stronger form of (II2) holds: namely that

(S.S.I)
(:3" c, 0: E n)(T, c, 0: distinct )(,llM'L.(c, ,; 0:) 1\ (M'L.(o:; c) n M'L.(o:; ,) = ¢Y».

Let r := M'L.(o:, {3; ,), II := M'L.(o:; {3), both of which exist by (HI) and (H2).
If, (j: II, then ME(o:; ,) n M'L.(o:; {3) =
¢Y and M'L.({3, ,; 0:) does not exist. So,
put f := {3, and (S.S.I) follows. Suppose on the other hand that, E II. In that
case, II ::> r. Let E E r.!\(II U {{3}). The element c exists by (HI). Suppose
A := M'L.(c, ,; IX) exists. Then c f. II U A and, E II\A. Therefore, by the syzygetic
property of E, and t.he fact. that II and A are chains of translates of 'L., we have
A C II. Hence;3 f. A, a cont.radiction to the assumption that M'L.({3,,; 0:) does
not exist. Therefore, ME(E, ,; 0') does not exist. Now let II' := Nl'L.(o:; c). Then
II' nIl = ¢Y. Thus (S.8.1) holds.
Now, define B(o:; {3, ,) if and only if

(,llg E C)(o: E 'L.g 1\{3" f. 2:;g)

(8.S.2)
I\(,llg E C)({3, , E 'L.g 1\ 0: (j: Eg).

Note that if (8.8.1) holds for " ( and 0:, then by the definition of M'L.(o:; f) and
M'L.(o:; ,), we see that B(o:;" f). Thus the relation defined in (S.S.2) is not empty.
Clearly (El) in (2.2.3) holds. Assume H(o:;{3,,) and B({3;o:,,). Suppose 0: i- {3.
By primit.ivity of C, there exists 9 E C such that 'L.g contains IX and excludes {3 or
there exists 9 E G such that 2:;g contains {3 and excludes 0:. If the former condition
holds, then by the definition of B(o:; {3, ,), " 0: E 'L.g and (3 f. 'L.g-contradicting
B({3,; 0:, ,). The other case leads similarly to" ;3 E 'L;g and 0: f. 'Eg by B({3; IX, , ) - -
contradicting B( 0:;(1, ,). So, we can deduce that B( 0:; (3, ,) and B({3; IX, , ) both
hold if and only if Cy = ;i. Therefore condition (B2) holds.
Infinite Jordan Permutation Groups 183

Assume, with hope of a contradiction, that

(8.8.3) B(o:; (J, I) 1\ -,B(o:; 6, I) 1\ -,B(o:; (J, 6).

From (8.8.3h (that is, the second part of (8.8.3)), either 'Eg contains 0: and excludes
6 and I for some 9 E G, or 'Eg' contains 6 and I and excludes 0: for some g' E G.
If such a 'Eg exists, then by (8.8.3h, {J E 'Eg. Thus 0:, (3 E 'Eg and 6, I ¢ 'Eg.
But then if'Eh contains 0: and excludes {J and 6 for any h, we have that 0: E 'Ehn'Eg,
6 ¢ 'Eh U 'Eg and {J E 'Eg\'Eh. Then by syzygetic property of 'E, 'Eh C 'Eg. Since
I ¢ 'Eg, then I ¢ 'Eh. Thus 0: E 'Eh, and (3, I' 6 ¢ 'Eh. This contradicts (8.8.3h- So
there is no 'Eh containing 0: and excluding {J and 6. In that case, by (8.8.3)a, there
exists some 'Eh' containing (3 and 6 and excluding 0:. This implies (3 E 'Eh' n 'Eg,
0: E 'Eg\'Eh', 6 E 'Eh'\'Eg. Then by the syzygetic property of 'E, 'Eh' U 'Eg = O.
Hence I E 'Eh'. Therefore (3, I E 'Eh' and 0: ¢ 'Eh'-contradicting (8.8.3h. So,
(8.8.3) leads to a contradiction if (8.8.3)2 implies existence of 'Eg containing 0: and
excluding 6 and f.
Suppose on the other hand that (8.8.3h implies existence of 'Eg' containing
6 and I and excluding 0:. Then {J ¢ 'Eg' by (8.8.3h- Then, just as in the last
paragraph, any 'Eh that contains 0: and excludes (3 and 6 must be disjoint from
'Eg' by the syzygetic property of 'E, and thus must exclude (3 and I. But this will
contradict (8.8.3h. And any Eh' that contains {J and 6 and excludes 0: must contain
'Eg' properly, and so must contain I-contradicting (8.8.3h again. So, (8.8.3) leads
to a contradiction in this case too. With this contradiction, we conclude that (8.8.3)
does not hold. So (B3) holds.
We conclude therefore that B as defined above is a B-relation. It is clear from
its definition that B is G-invariant.
Reasons for Remark 8.7.2.
Let (A, Bo) be the discrete B-set which is regular of branching type (2,3), as
established by [3, Theorem 2.9.7]. This means that every point of A has branching
number 2 and every branch point of negative type (that is, a branch point not in
A) has branching number 3 (see diagram below).
184 S. A. Adeleke

Let G := Aut(A, B o), the group of all maps of A preserving the B-relation.
From [3, Theorem 29.7]' (G,A) is primitive. Pick a branch point of negative type
and let E be a branch at that branch point. It is easy to see that E is syzygetic
with respect to G. Furthermore, Bo is not a betweenness relation since the three
points in the diagram above demonstrate that (A, B) violates axiom (B4).
We now show that any G-invariant B-relation on A violates axiom (B4). Let
a denote the top element in the diagram above; let {3 be the element at the lower
left and "( the third. Suppose B1 is an arbitrary G-invariant B-relation on A. As
observed from the action of G preserving B o, the action of G {a,fj,'Y} on {a, {3, "(} is
isomorphic to the symmetric group S3. Thus, by condition (B2) of the B-relation in
Section (2.2.3), none of a, {3, "( is between the other two in B 1 . Thus -,B 1 (a; {3, "(),
and, ifthere exists 6 E A such that B1 (6; a, (3) /I. B1 (6; a, "(), then 6 of {3 and 6 of "(.
Suppose such a 6 exists, and in addition, that Bo(a; 6, (3). Extend the diagram
of Bo from a to 6. (This will be away from (3). Because the branching number in
(A, Bo) is 3, then it is easy to see that there exists an element y E G of order 2 such
that
ay = a, {3y = (3, 6yof 6 and 6y2 = 6.
From B 1 (6;a,{3) and condition (B3), we have B 1(6;a,6y)V B1(6;{3,6y). Since B1
is G-invariant, this means that an application of y to the last property yields

The last statement contradicts condition (B2). Thus -,Bo( a; 6,13).

Similarly -,Bo({3;6,a), -,Bo(a;6,"() and -,Bo("(;6,a) hold for any 6 satisfying
Bl(6;a,{3)/I.Bl(6;a,"(). But there is no 6 E A\{a,{3,"(} in the diagram of (A,Bo)
that satisfies the last four conditions simultaneously. So, such a 6 does not exist.
Thus B1 violates axiom (B4) and cannot be a G-invariant betweenness relation.
Proof of Lemma 8.7.3.
Suppose (H 1) in Lemma 8.7.1 holds but (H2) fails. Then

(Va, {3, "( E n)(a,{3,,,( distinct)

(8.8.4)
(,llME(a, {3; "()) -+ «,llME({3, "(; a)) /I. (,llME("(, a; (3))).

Suppose also that

(8.8.5) (3ao, {30, "(0 E n)(ao, {30, "(0 distinct )(,llME(ao, (3o; "(0)).

Let r := ME(ao;"(o) and s E (n\(r U {a})). Both rand s exist by (HI). By

(8.8.5), (30 E r. Now ME(ao; s) exists again because of (HI). Furthermore,
ME(ao; s) n r = ¢ from the definition of c and r. Hence, (30, 'Yo ~ ME(ao; 0);
so ME({3o, 'Yo; c) exists. By (8.8.4), we have that ME(c, {30; "(0) exists. Let
II := ME(c, (30; 'Yo). By (8.8.5), a E II. Thus, "(0 Ern II, ao E II\f, {30 E r\II
Infinite Jordan Permutation Groups 185

and c: ~ r u IT. Since IT and r are increasing chains of translates of E, it follows

easily that there exist translates E 1 , E 2, of E with El ~ r, E2 ~ IT such that
,0 E El n E 2, ao E E2 \E 1 , Po E El \E 2 , and c: ~ El U E 2 , a contradiction to the
syzygetic property of E. This contradiction shows that if (HI) holds and (H2) fails,
then (8.8.5) does not hold. Thus, ME(a,,8;,) exists for all distinct a,p and, in
E.
Proof of Remark 8.7.4. Either (1) use E directly and the exact proof of [3,
Theorem 34.6] to prove Remark 8.7.4 or (2) note that O\E is syzygetic if E is, and
with O\E replacing E, condition 34.6.1 in [3] becomes condition (8.7.3.1) above.
Then, Theorem 34.6 proves the result above.
Proof of Lemma 8.7.5.
If E or O\E is hypersyzygetic, then by [3, Theorem 34.4], G preserves an
upper semilinear order or a C-relation. So, assume that neither E nor O\E is
hypersyzygetic but that E is syzygetic.
If (HI) holds and (H2) fails, then by Lemma 8.7.3 and Remark 8.7.4 above,
there is a G-invariant D-relation.
If (HI) above fails, then we have two cases:
Case 1: (8.8.6) (:la,,8 E O)(a,,8, distinct)(Vg)(a E Eg -+ ,8 E Eg);
Case 2: ME(a;,8) exists for all distinct a and ,8 but ME(ao;po) = O\{ao} for
some ao and ,80.
If Case 2 holds, then ME( ao; 6) = 0\ {ao} for all 6 distinct from ao. By
transitivity of G, ME( a;,8) = 0\ {a} for all distinct a and,8. As a consequence

(8.8.7) (Va,,8,, E O)(a,,8,, distinct )(3g E G)(,8" E E Aa ~ E).

Now if (8.8.7) holds, then by [3, Theorem 34.6]' G preserves a D-relation on O.

On the other hand, suppose Case 1 holds. Then by [3, Lemma 34.2]' there is a
G-invariant partial order::; on 0 defined by

a ;:::,8 if and only if (8.8.6) holds.

We shall now show that, in fact, (8.8.6) induces a G-invariant B-relation on O.

Define B(a;p, ,) if and only if (8.8.2) above holds. Then B becomes a G-invariant
B-relation as shown above in the proof of Lemma 8.7.1. Moreover, if a> ,8 > , in
the G-invariant partial order defined above, then it follows easily that B(P; a, ,).
Thus, the B-relation is not empty.
With Lemma 8.7.1 above, the proof of Lemma 8.7.5 is complete.

9 Jordan Groups on a Linear Order

9.1.
Let (G, 0) be a primitive Jordan group with a given G-invariant linear order::; on
O. Clearly, (G, 0) is simply primitive. And true, Theorem 7.1 shows that maximal
186 S. A. Adeleke

Jordan sets or minimal increments of Jordan sets induce either a G-invariant lin-
ear order or a nonlinear semilinear order through property (5.9.3), or a C-relation
through property (5.9.4) or (7.1.1). But those induced G-invariant relations may
be different from the linear order::; given a priori on Q. This section describes the
possible Jordan sets in (G, Q). It is obvious that in some cases, initial segments
and final segments of (Q,::;) are Jordan sets. But there are cases where they are
nol. In the latter possibility, every Jordan set is open, convex and bounded (that
is, bounded below and above), and even then not all open, convex, bounded sets
are Jordan sets. The ones that are Jordan sets can be described using a C-relation
or a nonlinear semilinear order. Concrete examples of cases with C-relation and
semilinear order are given in section (9.4.4) and (9.4.5) below. Theorems 9.3.1 and
\).3.2 (i) are joint work with Peter Neumann.
Remark 9.2. The following example is typical of the groups described in this
section. The detailed description is in section (9.4.4) below. Let A(:2, ~o, +) be the
upper semilinearly ordered set where each maximal chain is isomorphic to :2, and
where at each branch point, the set of branches is isomorphic to lQl with its usual
order. The notation A(:2, ~o, +) is from [3, section 5]. Let n be the collection of
all piecewise linear maximal chains in A(:2, ~o, +) using again a term in [3, section
5]. It is not hard to see that when the orders on the sets of branches at branch
points are considered together, they induce a linear order on the piecewise linear
maximal chains. Moreover, the set of maximal chains with lower sections in any
one branch at a branch point is convex in the induced linear order. This is the same
order that one gets if one, in a way, projects down the set Q onto a line. Let G
be an automorphism group of n that preserves the associated C-relation and the
induced linear order 011 o. One ends up having a linearly ordered set (Q,::;) with
a C-relation on rl, and a Jordan group (G, Q) with G preserving both < and the
C-relation.

9.:3 Results.
In the following theorems, let (G,O) be a primitive Jordan group with (G, Q) ~
(Aut(Q, ::;), Q) for some given linear order::; on Q.
Theorem 9.3.1 If (G, Q) is order 3-tmnsitzve wzth respect to ::;, then each max-
imal Jordan set containing one element and excluding another zs ezther an initial
segment or a final segment of (Q, ::;). Each maximal Jordan set then induces a
G-invariant lmear order on Q which is the same as ::; or the reverse of it.

Remark. Usual arguments show that (G, Q) in Theorem 9.3.1 is then order
n-t.ransitive for all n.
Theorem 9.3.2 Suppose (G,O) IS order 2- lmnszllve but not order 3-lransitive
with respect to ::;.
(z) Then every Jonlan set is an open bounded convex set in Q but not every
open bounded convex set is a Jordan set. Every maximal Jordan set vzolates property
Infinite Jordan Permutation Groups 187

(5.9.3) and satisfies property (5.9.4). So, it induces, as in the proof of Theorem 7.1,
a G-invariant relation which is neither linear nor semilinear but rather a C -relation.
(ii) Furthermore, G is transitive on the set of nodes of the C-relation. If the number
of branches at a node, and therefore at every node, is greater than 2, then it is
infinite, and the linear order on D induces a dense linear order on the set of branches
at each node. On the other hand, if the number of branches at every node is 2, then
between any two comparable nodes, there is another.
Theorem 9.3.3 Suppose (G, D) is not order 2-transitive. Then, again, every
Jordan set is an open bounded convex set but not vice-versa. Any maximal Jordan
set A that satisfies property 5.9.3 does not induce through the property an invariant
linear order but a nonlinear semilinear order. The other possibility is for a maximal
Jordan set to induce a G-invariant C-relation through property (5.9·4)·
Result 9.3.4. An example is indicated in section 9.4.4 below of the C-relation
in Theorem 9.3.2 which has an infinite number of branches at each node (that is,
branch point), but for which the set of nodes on each maximal chain is isomorphic
to Z.
Result 9.3.5. Section 9.4.5 below contains an example of the case in Theorem
9.3.3 where G preserves a nonlinear semilinear order on D.
Moreover, the example is neither weakly doubly transitive nor a right regular
representation of a subgroup of JR, nor periodic. Thus it answers in the negative a
primitivity question in [18, page 249].
9.4 Proofs.
9.4.1. Proof of Theorem 9.3.1. Let r be a Jordan set in D. Then clearly, r
is convex. Suppose r is bounded with a E r, ;3 tt. r, and a < ;3. Since Gco\r)
is transitive on r, then a is not a minimal element in r. So, r contains some 6
with 6 < a. Let J.l E O\r with J.l < 6. By order 3-transitivity of G, there exists
x E G with (6, a, ;3)x = (J.l, a, ;3). Thus J.l E M J(;3; a), the maximal Jordan set in
D containing a and excluding;3. Similar arguments show that if a < 1/ < ;3 then
1/ E M J(;3; a). Hence M J(;3; a) is the initial segment of (O,~) with least upper
bound;3. We conclude, using similar arguments, that in any case, every maximal
Jordan set A is an initial segment or a final segment of (D, ~). It then follows that
Ag ~ A or Ag 2 A for every 9 E G. In addition, A satisfies property (5.9.3). Hence
A induces through property (5.9.3) and in the manner of proof of Theorem 7.1 a
G-invariant linear order on D. From the description of A above, the linear order
equals the given order ~ in the hypothesis of this theorem or the reverse of it.
9.4.2. Proof of Theorem 9.3.2(i). Let r be a Jordan subset in D. Then again, r
is convex with respect to the linear order ~ on D. If r is unbounded, then by usual
arguments, repeated below, G is order 3-transitive. For suppose r is unbounded
below. Then order 2-transitivity of G implies that for any a E D the set 1::" of
elements less than a is a Jordan set. Given al < a2 < a3 and a~ < a~ < a~,
188 S. A. Adeleke

we can map (a2' (3) to (a~, a~) by order 2-transitivity and can use the elements of
GCO\E I) to adjust our map to take (al' a2, (3) to (a~, a~, a~). This will mean that
"'2
G is order 3-transitive, a contradiction to the hypothesis. A similar contradiction
results if G has Jordan sets which are unbounded above. So, r must be bounded.
It follows also easily from order 2-transitivity that :::; is dense. Since G(O\!,) is
transitive on r then r cannot contain a maximal or a minimal element. So r is
open. Thus r is an open convex bounded set with respect to :::;, a dense linear
order.
Suppose a E rand (3 ff. r. Define A := M J«(3; a). From the last paragraph A
is a bounded interval in (n, :::;). Suppose a < (3. An argument similar to the one
below treats the case when a > (3. From boundedness of A, there exists 'Y E n\A
such that 'Y < a. By order 2-transitivity of (G, n), G contains some element 9
satisfying ('Y,a)g = (a,(3). Thus from 'Y ff. A and a E A, we deduce a r;. Ag and
(3 E Ag. Thus property (5.9.3) fails to hold for the maximal Jordan set A containing
r.
By the sketch proof in Section 7.1, the two remaining possibilities are property
(5.9.4) or (7.1.1) holding. In fact A always satisfies property (5.9.4). For if otherwise,
then by arguments similar to those in Section 5.9.2, there exists 9 E G such that A,
Ag are typical and 9 interchanges (3 and (3g-that is, «(3, (3g)g = «(3g, (3). But such a
9 cannot occur here where 9 preserves a linear order. Thus A induces a G-invariant
C-relation through property (5.9.4).
Proof of Theorem 9.3.2 (ii). The C-relation on n means that the elements of n
behave like some maximal chains in an "inverted tree" , or equivalently, in an upper
semilinearly ordered set (see [2], [3], [4] or [6]. One of the 4 possible configurations
of C(a; (3, 'Y) in this case is shown.
Definitions. For a, (3 E n with a i (3, define

N{a,(3}:= bEn 1'C('Y;a,(3nu {a,(3}

BR(a; (3) := bE nlC(a; (3, ,n·

C(a;I3,Y)

From the definitions, a, (3 E N {a, (3} and (3 E B R( a; (3). We notice that N {a, (3}
corresponds to a node while BR(ai(3) corresponds to the branch at N{a,(3} that
contains (3. By order 2-transitivity, G is transitive on the set of all nodes of n. For a
similar reason, we note that when a node is mapped to another node, the branches
Infinite Jordan Permutation Groups 189

at the node are mapped in a bijective manner to the branches at the new node.
Note also that since, from the proof of Theorem 9.3.2(i), the C-relation is induced
by A which satisfies property (5.9.4), then BR(a; {3) is a Jordan set. So, BR(a; {3)
is convex. Consequently, ~ induces an order on the set of branches at each node.
Suppose the branches at a node are more than 2 in number. In particular, let
lEN {a, {3} with a < {3 < I in the linear order on O. By order 2-transitivity of
G, we can translate the pair (a, {3) to (a, I)' Since N {a, {3} is then fixed and the
branches are permuted amongst themselves, we see that there is a branch between
a and {3. Hence the order on the branches induced by linear order ~ on 0 is dense.
By taking (a, {3) to ({3, I), we notice that within a node, there is no minimal or
maximal branch either. Furthermore, the order on the branches at a node is order
2-homogeneous.
Suppose a node has only two branches. Since G is transitive on the set of
all nodes, then every node has two branches. Suppose there is no node between
N {a, 11} and N {a, 12} with the latter being lower than the former in the semilinear
ordering. Suppose a < 12. (The reader may want to draw a diagram here). Now
BR(,1; a) being Jordan, there exists Z E GCO\BR(-;l;a)) such that 12Z = Q. In that
=
case az < 12Z a. Since 12 f. a and QZ < a, then N{az, a}:f. N{a"2}. We then
have two distinct maximal nodes within BR( 11; a) - a contradiction to the upper
semilinear ordering on the set of nodes. Hence, between N { a, 1d and N { a, 12},
there is another node. This conclusion holds by similar arguments if we assume
12 < a. Hence, the nodes are dense with respect to the semilinear ordering on
them.
9.4.3. Proof of Theorem 9.3.3. Let A := M J({3; a). Suppose, with hope of
a contradiction, that A induces a linear order on 0 through property 5.9.3. Then
Ag n A :f. ¢ for any 9 E G. Moreover, from property (5.9.3), any translate of A that
contains one element of ~(A) contains A properly and thus contains all elements
of ~(A). Thus ~(A) becomes a block for (G,O). Since (G,O) is primitive, then
~(A) = {{3}. From Ag n A f. ¢ for any 9 E G, we deduce that A is not bounded
with respect to the a priori order ~ on 0; otherwise by moving {3 suitably, some x
=
in G will satisfy Ax n A ¢. Since ~(A) =
{{3}, we can now assume without loss
of generality that A is the collection of all elements of 0 less than {3. It is then easy
to show that G is order 2-transitive on (0, ~). This contradicts the hypothesis of
the Theorem. Thus, no maximal Jordan set can induce an invariant linear order
through property 5.9.3. By the arguments at the close of Theorem 9.3.2(i), we
then see that every maximal Jordan set must satisfy property 5.9.4 which yields a
G-invariant C-relation on O. Hence the theorem holds.
9.4.4. Example on Result 9.3.4.
Let
A'-
{(On1Q1 n 2Q2 ... n k-1qk-1 n kl k E N, Qi E iQ\{O}, i = 1, ... , k, n1 > ... > nk E iZ}.
Consider the following order ~. on A :
190 S. A. Adeleke

(On1q1 n 2q2 ... n k -1 qk -1 nk) ~s (Om1Plm2P2 ... m r -1 Pr -1 m r )

if and only if:
(i) k ::; r,
=
(ii) ni mi and qi = Pi for i = 1,2, ... , k - 1, and
(iii) nk ~ mk·
Remarks. (A, ::;8) becomes an inverted tree with an infinite number of branches
at each node, that is, a homogeneous upper semilinear order with the set of elements
of A on each chain having order type &::. Then A equals A(&::, ~o, +) using the
notation in [3, section 5].
Description of a linear order on some chains of (A, ::;.).
Define

0'-
{(OnlQln2q2 ... nkqk) IkE NU {O},Qi E Q\{O},i = 1, ... ,k,nl > ... > nk E &::}.

In [3, section 5], 0 is called the set of piecewise linear maximal chains in A. We can
view (0) as the vertical maximal chain; (OnlQ1n2Q2" . nkQk) as the maximal chain
in the qkh branch at node (Onl q1 ... nk) which is straight after the stated node
(that is, it contains no node of the form (Onlql ... nkqknk+1qk+1nk+2) or lower). If
W2 := (On1Q1 ... nkqk), and WI := (Om1P1 ... mrPr), define:
W2 >1 WI if and only if
WI is an initial segment of W2 and qr+l > 0
or W2 is an initial segment of WI and Pk+l < 0
or for some j ~ 1, qi = Pi for i = 1, 2, ... ,j - 1,
ni = mj for i = 1,2, ... ,j, and qj > Pi;
= =
or for some j ~ 1, qi Pi, nj mi for i =
1,2, ... , j - 1,
and nj > mj and qj > 0;
= =
or for some j ~ 1, qi Pi, ni mi for i =
1,2, ... ,j - 1
and nj < mj and Pj < O.

For example, in the diagram, we have 11 <I 12 <I 13 <I (0) <I 14 <I 15. We
can verify that relation::; becomes a dense order 2-homogeneous linear order on
 Infinite Jordan Permutation Groups 191

Q. The relation also induces, on the set of all branches at a node, a linear order
isomorphic to the order on Q.
Let G be the subgroup of Aut(A, ::Ss) which preserves on each node the linear
order amongst the branches at the node and which fixes Q setwise. One can verify
that G ~ Aut(n, ::S1) and is an order 2-transitive Jordan group which is not order
3-transitive.
9.4.5. Example on Result 9.3.5. Consider the case (G, Q) in Theorem 9.3.2 (ii)
where the number of branches at each node is exactly 2 and where the set of nodes
on each maximal chain is dense. From this denseness, each branch BR(f3; a) in Q is
not only a Jordan set but an order 2-transitive Jordan set with respect to the order
::s
on the set Q of maximal chains.
Let II be the collection of all nodes in the C-relation on Q. It is not difficult
to see that II is an upper semilinearly ordered set. The order 2-transitivity of the
Jordan sets in (G, n) now shows that the set E of all nodes contained in any branch
below a given node is a Jordan set in II. In fact, denseness of II relative to the
upper semilinear order on it implies that E is a primitive Jordan set. It thus follows
that (G, II) is a primitive Jordan group. Moreover, the linear order ::s
on Q now
induces a linear order ::s.
on II as follows. To make the definition easier, first note
that if 7r E II, the representation 7r = N{w,w'} of 7r for w,w' E Q is not unique.
Secondly, if 11"1, 7r2 are comparable relative to the upper semilinear order on II, then
there exist distinct W1,W2,W3 E Q such that 11"1 = N{Wl,W2}, 11"2 =N{W2,W3} and
WI < W2 < W3 or WI > W2 > W3 where ::sis the linear order on Q. And if 7r1, 7r2
are incomparable, then there exist distinct W1,W2,W3,W4 such that 7r1 = N{W1,W2},
11"2 = N {W3, W4} and WI < W2 < W3 < W4 or WI > W2 > W3 > W4. SO, we can state
the definition of ::s.
in general as:

if and only if
there exist 11,/2, ... , 1m E Q such that:
11 < 12 < 13 < ... < 1m,
N{a1,ll:2} = N{fl,/2}, and
N {f31, f32} = N {fm-1, 1m}.
The form of this definition makes the proof that <. is a linear order very simple.
Clearly, G is not order 2-transitive with respect to order ::S. on II since two
comparable elements under ::Sa cannot go into incomparable elements under ::Ss.
But we shall show even more; namely that (G, Q) is not weakly doubly transitive
(see (2.2.11) above for definition).
Note that (II, ::s.) contains a subset of the following form:
192 S. A. Adeleke

In the diagram, a, {J, 'I, {j refer to maximal chains, and 11"1,11"2,11"3 refer to nodes. Note
that in the diagram, 11"1, 11"2, 11"3 are comparable nodes under ~s, and that

a < {J < 'I < 6 III (n,~)

11"1 = N{a,{J},1I"2 = N{{J,'I} and 11"3 = Nb,6}.
So, 11"1 <* 11"2 <* 11"3. But there is no element in e that fixes node 11"1 and moves
11"2 to 11"3. In fact, observe that if 11" E n such that 11" ~* 11"3, then 11" is 11"1. Thus
for any such 11", there is no element in e that fixes node 11"1 and moves 11"2 to 11".
Hence, (e, II) is not weakly doubly transitive using the definition in (2.2.11). It
is also easy to see that (e, n) is not periodic (see 2.2.11 for definition) since any
z satisfying the second condition of periodicity cannot fix setwise each Jordan set
in (e, II), and therefore cannot satisfy the commutativity condition of periodicity.
The existence of Jordan sets also shows that (e, II) is not a regular representation
of any subgroup of JR. With this, the proof of Result 9.3.5 is complete.

References
[1] S. A. Adeleke and Peter M. Neumann, Infinite bounded permutation groups, J.
London Math. Soc., to appear.
[2] S. A. Adeleke and Peter M. Neumann, Primitive permutation groups with prim-
itive Jordan sets, J. London Math. Soc., to appear.

[3] S. A. Adeleke and Peter M. Neumann, Relations related to betweenness: their

structure and automorphisms, preprint.

[4] S. A. Adeleke and Dugald Macpherson, Classification of infinite primitive Jor-

dan permutation groups, Proc. London Math. Soc., to appear.
Infinite Jordan Permutation Groups 193

[5] S. A. Adeleke, M. A. E. Dummett and Peter M. Neumann, On a question

of Frege's about right-ordered groups, Bull. London Math. Soc. 19(1987),
513-521.
[6] S. A. Adeleke, Semilinear towers of Steiner systems, J. Comb. Theory Series
(A), to appear.
[7] S. A. Adeleke, Examples of irregular infinite Jordan groups, preprint.
[8] Peter J. Cameron, Transitivity of permutation groups on unordered sets, Math.
Zeitschrift. 148(1976), 127-139.
[9] Peter J. Cameron, Orbits of permutation groups on unordered sets, IV: homo-
geneity and transitivity, J. London Math. Soc. (2) 27(1983),238-247.
[10] Peter J. Cameron, Some tree-like objects, Quart. J. Math. Oxford 38(1987),
155-183.
[11] G. Cherlin, L. Harrington, and A. H. Lachlan, ~o-categorical, ~o-stable struc-
tures, Annals of Pure and Applied Logic 28(1985), 103-135.
[12] J. A. Covington, A universal structure for N -free graphs, Proc. London Math.
Soc. (3) 58(1989), 1-16.
[13] Manfred Droste, Structure of partially ordered sets with transitive automor-
phism groups, Memoirs American Math. Soc. 57(1985), No. 334.
[14] Manfred Droste, The normal subgroup lattice of 2-transitive automorphism
groups of linearly ordered sets, Order 2(1985), 291-319.
[15] M. Droste, W. C. Holland and H. D. Macpherson, Automorphism groups of
semilinear orders, I, II, Proc. London Math. Soc. (3) 58(1989), 454-478;
479-494.
[16] David M. Evans, Homogeneous geometries, Proc. London Math. Soc. (3)
32(1986), 305-327.
[17] D. Giorgetta, Dr. Phil. Dissertation, E. T. H., Zurich.
[18] A. M. W. Glass, Ordered Permutation Groups, London Math. Society Lecture
Notes Series 55, Cambridge University Press, Cambridge, 1981.
[19] Camille Jordan, Sur la limite du degri des groupes primitifs que contiennent
une substitution donnie, J. Reine und Angew. Math. (Crelle's J.) 79(1875),
248-258 (::::Oeuvres de Camille Jordan Vol. I (Paris 1961),485-495).
[20] W. M. Kantor, Homogeneous designs and geometric lattices, J. Comb. Theory
Series A 8(1985), 64-77.
[21] William M. Kantor, Jordan groups, J. Algebra 12(1969), 471-493.
194 S. A. Adeleke

[22] H. D. Macpherson, A survey of Jordan groups, in Automorphism Groups of

First Order Structures, eds. R. W. Kaye and H. D. Macpherson, Oxford Uni-
versity Press, Oxford (1994), 73-110.
[23] H. D. Macpherson, Extensions of Jordan groups, preprint.
[24] T. P. McDonough, An infinite version of Marggraff's theorem, Quart. J. Math.
Oxford (2) 32(1981),173-179.
[25] Peter M. Neumann, The lawlessness of groups of finitary permutations, Archiv
Math. (Basel) 26(1975), 561-566.
[26] Peter M. Neumann, The structure offinitary permutation groups, Archiv Math.
(Basel) 27(1976),3-17.
[27] Peter M. Neumann, Some primitive permutation groups, Proc. London Math.
Soc. (3) 50(1985),265-281.
[28] Peter M. Neumann, The classification of some infinite Jordan groups, preprint.
[29] D. Segal, Normal subgroups of finitary permutation groups, Math. Z. 140
(1974),81-85.
[30] James Wiegold, Groups of finitary permutations, Arch. Math. 25(1974)' 466-
469.
[31] H. Wielandt, Unendliche Permutaiionsgruppen, Lecture notes, Tiibingen, 1960
(reprint: York University, Toronto, 1967).
[32] B. I. Zilber, Finite homogeneous geometries, in Proc. of the 6th Easter Confer-
ence on Model Theory, eds. B. I. Dahn and H. Wolter, Humboldt Universitat
zu Berlin (1988), 186-208.
 The Separation Theorem for Group Actions

Cheryl E. Praeger

University of Western Australia

Nedlands, W.A., 6907
Australia

1 Introduction

Let G be a group of permutations of a set n, and let r and ~ be (not necessarily

distinct) subsets of n. Under what conditions on G, or on r and ~ is it possible to
separate r from ~ by an element of G, that is to have P n ~ = 0 for some g E G?
In 1976 Peter M. Neumann [25, Lemma 2.3J proved that every finite subset can be
separated from itself in this way provided that all the G-orbits in n are infinite.

Theorem 1.1 (Separation Theorem) If G is a group of permutations of a set n

such that all G-orbits in n are infinite, then for each finite subset r of n there is
an element g E G such that f9 n r = 0.

This fundamental result for group actions was shown in [25J to be equivalent
to an earlier result of B. H. Neumann [24, Lemma 4.1J about covering an abstract
group by a finite number of cosets of proper subgroups.
This chapter presents an account of the Separation Theorem for group actions.
We shall discuss various improvements and generalisations of Theorem 1.1 which
have been proved in an attempt to understand its full power. We shall also sur-
vey several diverse applications to problems about group actions, abstract group
structure, and even an unsolved problem about combinatorial designs.
The contrapositive of the assertion made by Theorem 1.1 is the following. If
there is a finite subset r <;:; n such that, for all g E G, Ifg \ fl < If I, then G has at
least one finite orbit in n. It is in this form that the Separation Theorem is usually
applied, and it is from this form that questions most readily arise. How large is
the finite orbit? Is its size dependent on the size of the finite set f? Is any similar
assertion possible when r is infinite? It is from questions such as these that many of
the variations, improvements and generalisations of the Separation Theorem have
ansen.
For a finite subset r of n, the movement of r under the action of a group G of

195
W. C. Holland (ed.), Ordered Groups and Infinite Permutation Groups, 195-219.
© 1996 Kluwer Academic Publishers.
196 Cheryl E. Praeger

permutations of n is defined as

move (f) := max

9EG
Irg \ rt,
and l' will be said to have restricted movement if move (r) < Ifl. Thus an equivalent
version of the Separation Theorem is:
Theorem 1.1' Let G be a group of permutatwns of a set n. If there exists a
finite subset l' of n wlfh restricted movement, then G has at least one finite orbit
in n.
Sections 2 and 5 are concerned with specifying upper bounds on the length of
the smallest G-orbit in terms of parameters such as the size and movement of f.
Under certain conditions we can extend the definition of restricted movement
to infinite subsets: if l' is any subset of n such that, for all g E G, 1'9 \ I' is finite,
and {Ira \ I'll 9 E G} is bounded, then we define the movement move (f) of 1', as
above, as the maximum of 11'9 \ rt over all 9 E G, and we say that I' has restricted
movement if move (f) < Irt. Of course if move (r) is defined for an infinite subset
1', then move (f) will be finite and hence less than 11'1, so r will have restricted
movement. If n is infinite then every cofinite subset of n has restricted movement,
so there are many infinite subsets with restricted movement. More generally, if ~
is a G-invariant subset of n (that is, if 1";9 ~ 1"; for all 9 E G) then move(~) = 0
and ~ has restricted movement. Section 3 reports on a study in [8] of the structure
of subsets with restricted movement: it is shown that a subset with restricted
movement m differs from a G-invariant subset by a set of size bounded above by
a function of m. In Section 4 we apply this result, together with a result from
Section 2, to obtain a version of Theorem 1.1' for arbitary subsets with restricted
movement.
Consider the situation in which, for a permutation group G on n, a large
number of subsets of n have movement bounded above by a positive integer m.
For example, if all subsets of n have finite movement at most m then both the
number of nontrivial G-orbits and their lengths are bounded above by functions of
m. Similarly if, for some integers m and k such that 0 < m < k ~ In!, all k-element
subsets of n have movement at most m, then the number of nontrivial G-orbits
and their lengths are bounded above by functions of m and k. These results are
discussed in detail in Section 6.
The Separation Theorem has been used in some surprising ways. In Section 7
we give a brief account of the way it was used in solving a problem about abstract
groups. For k a positive integer, a group G is said to have the small squaring
property SS(k) if, for all k-element subsets ]{ of G, the set of all products gh,
for g, h (not necessarily distinct) elements of ]{, has size strictly less than k 2 . An
abelian group G has the small squaring property SS (k) for all k, 2 ~ k ~ IGI. A
version of the Separation Theorem was used to show that an infinite group with
property SS(k) for some k is not far (in a certain sense) from being abelian.
 The Separation Theorem for Group Actions 197

Not surprisingly for such a fundamental result as the Separation Theorem,

special cases of it have been proved independently for particular kinds of group
actions. The most commonly used quantitative version of the Separation Theorem
[2] (stated in Section 2 as Theorem 2.1) was also proved in [18, Lemma 7] in the
case of the symmetric group on a set E of size n acting on the set n of m-element
subsets of E. This version has been applied to problems in both Graph Theory
and the theory of Combinatorial Designs, and we shall discuss these applications in
Section 9. Our account in Section 9 makes use also of a probabilistic version of the
Separation Theorem which will be presented in Section 8.

2 Quantitative versions of the Separation Theorem

This section contains a discussion of various versions of the Separation Theorem

which specify upper bounds on the size of the finite orbit in the conclusion of
Theorem 1.1'. The first, and most useful result, was proved by B. J. Birch, R.
G. Burns, S. Oates Macdonald and P. M. Neumann [2, Theorem 2] in 1976 and
gives an upper bound in terms of the size of the finite set r. Their most general
result is proved for the separation of possibly different finite subsets r, Ll. (In
the statement we use the notation move (f) for the movement of a finite subset r
defined in Section 1.)
Theorem 2.1 [2, Theorem 2] Let G be a group of permutations of a set 0, and sup-
pose that there are finite subsets r, Ll of 0 such that ru n Ll f:. 0 for all g E G. Then
there exists a G-orbit in 0 of length at most Irl·ILlI. In particular, if move (f) < Irj,
then some G-orbit in 0 has length at most Ir12.
Theorem 2.1 clearly gives more precise information than the Separation Theo-
rem, and the Separation Theorem follows from Theorem 2.1. However there is no
known proof of Theorem 2.1 which does not rely on the Separation Theorem.
Question 1 Is there a proof of Theorem 2.1 which does not use the Separation
Theorem 1.1?
The upper bound in Theorem 2.1 is the best possible bound in the case of
separating different subsets rand Ll. For example, if 101 = km and G = Sic wr Sm
is the subgroup of the symmetric group Sym (0) preserving some partition P of 0
into m subsets of size k, and if r is one ofthe subsets ofP while Ll consists of exactly
one point from each subset ofP, then rg n Ll '# 0 for all g E G. However, if f = Ll,
then it is implicit in the proof of Theorem 2.1 that the bound Irl 2 is not sharp,
and this was formally recorded for transitive group actions in [8, Proposition D].
Moreover if G is intransitive on 0 then a better bound, namely (Ir/ - 1)2, was
obtained in [29, Corollary 1.5]. (In Section 5 we discuss in more detail the case of
intransitive group actions.) Although the bound /f/ 2 in Theorem 2.1 is not sharp
for transitive groups G, it is not far from best possible, as the following family of
examples shows.
198 Cheryl E. Praeger

Example 2.2 Let k - 1 be a prime power, let [2 be the set of points of the
Desarguesian projective plane PG 2(k - 1), and let r be the set of points on a line
of PG 2(k - 1). Then G := PGL (3, k - 1) is transitive on [2, 1[21 = k 2 - k + 1, and
f is a k-element subset with move(r) < k.
The set E := {PI g E G} forms the set of (hyper)edges of an intersecting
hypergraph having all edges of size k, that is a k-uniform hypergraph. According
to Z. Fiiredi (in [14, p.158]) the conjecture that the union of the edges of such a
hypergraph contains at most k 2 - k + 1 points was made by P. Erdos and by B.
Bollobas, and was first proved by L. Lovasz in 1975. A second proof was given
in 1981 by Fiiredi [14, Corollary 4] using the theory of fractional matchings and
fractional transversals in hypergraphs. Fiiredi's proof also shows that any example
attaining the bound k2 - k + 1 comes from a projective plane of order k - 1. A
proof of these facts involving only elementary counting arguments, and using the
language of tactical configurations can be found in [27, Corollary 1].
Theorem 2.3 Let G be a group of permutations of a set [2, and suppose that f is
a k-element subset of [2 with move (f) < k. Then some G-orbit in [2 has length at
most k 2 - k + 1. Furthermore, iJ the shortest G-orbit [2' intersecting f nontrivially
has length k 2 - k + 1, then f ~ [2' and the G-translates of f are the lines of a
projective plane with point set [2'.

Extra information about the action of the group G on 0, and the corresponding
bounds on the minimal length of G-orbits in Theorem 2.1, can be obtained by taking
into account the value of the following parameter describing the separation of two
subsets.
For a permutation group G on [2, if f and ~ are subsets of [2 such that P n ~
is finite for some g E G, we define the separation defect of f and ~ under the action
of Gas
sepdef (f,~) := min Ifg n ~I,
gEG

so that r and ~ can be separated by G if and only if sepdef (r,~) = o.

Theorem 2.4 [30, Theorem 4] Let G be a group of permutations of a set [2 and
suppose that, Jor some pair of finite subsets f, ~ oJO, sepdef (r,~) =
m > O. Then
there exists a G-orbit in [2 of length at most Irl . I~I/m.
As in the case of Theorem 2.1, the bound in Theorem 2.4 is sharp in the case of
separating different subsets r and ~ whenever m = sepdef cr,~) divides Irl ·I~I.
I am grateful to Avinoam Mann for pointing out the following generalisation of
my earlier class of examples meeting the bound of Theorem 2.4. Write m = kl
where k divides Irl and I divides I~I. Let 0 be a set of size Irl . I~I/m and
let G = 5WI/k wr 5\6.1/1 be the subgroup of the symmetric group Sym([2) which
preserves a partition P of 0 into I~I/l subsets of size Ifl/k. If f is the union of k
parts of P and ~ consists of I points of each of the parts of P, then each image of
The Separation Theorem for Group Actions 199

f under an element of G meets d in exactly m points so that sepdef (f, d) = m.

From the proof of Theorem 2.4 in [30, Section 5], it follows that, in any example for
which the bound If I· Idl/m is attained, f, d must both be subsets of a G-orbit of
length Ifl 'Idl/m, and the cardinality of f9 n d must be equal to m for all g E G.
It would be interesting to know if the examples above are the only examples of
transitive groups G of degree Ifl·ldl/m for which there are subsets f, d such that
sepdef (f, d) = m.
However, the upper bound provided by Theorem 2.4 in the case where f = d,
namely Ifl 2 /m, is not best possible. This can be seen when m = 1 by applying
Theorem 2.3. A better upper bound which can be attained for certain arbitrarily
large values of the relevant parameters, can be deduced from a result of Frankl
and Fiiredi [12, Theorem 5.4] about intersecting hypergraphs. Alternative proofs
have been given in [9] (using the theory of Bose Mesner algebras of association
schemes) and in [27, Corollary 2] and [30, Theorem 1] (using elementary counting
arguments). The result is the following. (A 2-(v, k, A) design consists of a set 0 of
v points together with a collection of k-element subsets of 0, called blocks, such
that each pair of points is contained in exactly A blocks; such a design is said to be
symmetric if there are exactly v blocks.)
Theorem 2.5 Let G be a group of permutations of a set 0 and let m and k
be positive integers with m < k. If there exists a k-element subset f of 0 such
that move (f) ::; m, then f intersects nontrivially a G-orbit in 0 of length at most
(k 2 - m)/(k - m). Moreover if the shortest G-orbit 0' intersecting f nontrivially
has length (k 2 - m)/(k - m), then f ~ 0' and the G-translates off form the blocks
'
of a symmetric 2-( \2~::: k, k - m) design with point set 0'.
It is quite straightforward using the techniques of [2] to obtain k 2 /(k - m)
as an upper bound in Theorem 2.5, and this was obtained in [8, Theorem 2.1] in
the case where G is transitive on O. If k = m + 1 we recover Theorem 2.3. Also
if m = 1 and the shortest G-orbit 0' meeting f has length k + 1 (the bound of
Theorem 2.5) then we have the trivial case in which f is the whole orbit 0' with
one point removed. For the other cases, that is, for 2 ::; m ::; k - 2, in the case of
equality in Theorem 2.5 the fact that k - m divides k 2 - m (or equivalently k - m
divides m 2 - m) provides a nontrivial restriction on the parameters.
Note that every symmetric 2-( v, k, A) design has the property that each pair of
distinct blocks has precisely A points in common (see for example [11, p. 57]). Thus
in Theorem 2.5, if the transitive action of G on 0' is equivalent to the action of G
on the points of some (in fact, any) point-transitive symmetric 2-(v, k, A) design,
and if f is a subset of 0' which forms a block of this design, then f has restricted
movement equal to k -..\ and the upper bound on 10'1 is attained. Typical examples
of such designs are the designs from projective geometries in Example 2.6 below.
Other examples of point-transitive symmetric designs are the Paley designs, defined
in [11, p.97], which have k = 2m - 1 where 4m - 1 is an odd prime power (see
Example 3.5 in the next section).
200 Cheryl E. Praeger

ExaIllple 2.6 Let q he a power of a prime, and n an integer, n 2: 2. Let n

be the set of points of the projective geometry PG n(q), let G = PGL (n + 1, q)
(or any subgroup transitive on 0), and let l' be the set of points on a hy-
perplane. Then k := Ifl = (qTl - l)/(q - 1), m := move(r) = qn-l, and
= =
101 (qn+l - l)/(q - 1) (P - m)/(k - m).

3 Subsets with restricted movement:

the closest invariant set

In [8] a study was made of the structure of (possibly infinite) subsets r of a set 0
which have restricted movement under the action of some permutation group G on
0, that is, subsets r such that 1'9 \ r is finite and of bounded size for all 9 E G so
that move (r) is defined, and such that move (1') < Ifl. The aim in [8] was to show
that such a set r must have symmetric difference with some G-invariant set 1; of
size bounded above by a function of move (r).
The suggestion that this might be the case came from a result of Leonid
Brailovsky [3] that a subset r with (restricted movement and with) move (r) = 1
must be equal to some G-invariant subset with either one point added or one point
removed. It follows from the definitions that a subset l' is G-invariant if and only if
move (r) = 0, and this is the case if and only if r has symmetric difference of size
o wit.h some G-invariant set, namely with r itself. In general, if r has size greater
than m and if r can be obtained from a G-invariant subset 1; by adding and/or re-
moving a total of at most m points, then r has restricted movement, move (r) ::::; m,
and the symmetric difference r 61; has size at most m: we shall call such subsets r
generic subsets with restricted movement at most m.
Certainly not all of the subsets r with restricted movement at most m (that
is move(r) ::; m) are generic. For example, when m is a prime power, it follows
from Example 2.2 that a line r in the Desarguesian projective plane PG 2(m) has
move (1') = m < WI = m + 1, and the closest PG L (3, m )-invariant subset to r is the
empty set 0. In [7] it was shown first that, if move (r) = 2, then r has symmetric
difference at most 3 from some G-invariant subset. Then the bound 3 was shown to
be best possi ble and in all cases meeting this bound the collection {r g I 9 E G} was
shown to come from one of a handful of possible geometrical configurations. These
results prompted the general study, and the best general result obtained in [8] is
the following. (For a real number x, ixl denotes the least integer greater than or
equal to x. Also, e is the base of natural logarithms.)
TheoreIll 3.1 [8, Theorems A, B, and Corollary 2.3] Lei G be a group of permuta-
tions of a set 0 and suppose that r ~ 0 has restricted movement with move (r) = m.
Then
(aJ there exists a G-invariant subset ~ ofn such that Ir6~1 < 2emi(ln2m)l, and
(b J l' has a proper, nontrivial intersection with at most 2m - 1 of the G-orbits.
(cJ Moreover, if G is transitive on 0 then either 11'1 : : ; 2m - 1 or 10 \ 1'1 ::; 2m - l.
 The Separation Theorem for Group Actions 201

The bounds in parts (b) and (c) are sharp for some values of m as Examples 3.2
and 3.3, respectively, show.
Example 3.2 Let m := 2r ;::: 2, and let G := Z;+I. Then G has exactly
2m - 1 = 2r +1 - 1 subgroups HI, ... , H 2m - 1 of index 2. Let OJ := {Hjg I g E G}
be the set of (two) Hj-cosets, for each i, and set 0 := UI<j<2m-1 OJ. Then G acts
faithfully on 0 by right multiplication with 2m - 1 orbits-OI, ... , 02m-1 of length
2. Note that each nontrivial element g E G lies in exactly m - 1 = 2T - 1 of the Hi
and therefore permutes nontrivially the remaining m = 2T of the OJ. Let f consist
of exactly one point from each of the OJ. Then Ifl = 2m - 1 and move (f) = m, so
f has restricted movement, and f has a proper nontrivial intersection with each of
the 2m - 1 orbits.
Example 3.3 An Hadamard design 'D is a 2 - (4m - 1, 2m - 1, m - 1) design, that
is'D consists of a set 0 of 4m - 1 points and a set of (2m - I)-element subsets of
0, called blocks, such that each pair of points lies in exactly m - 1 blocks (see [11,
p97] for this definition and for the proofs of various assertions made in the following
discussion). An Hadamard design 'D is a symmetric design, that is the number of
blocks is equal to the number 4m - 1 of points. Also each pair of distinct blocks of
'D intersects in exactly m - 1 points. Let G be the automorphism group of 'D, so
elements of G map blocks to blocks, and let f be a block of 'D. If g E G, then f g \ f
is either the empty set or has size m. Thus f has restricted movement, If I = 2m-I,
and move (f) = m under the action of G on O. If G were transitive on 0, then
the closest G-invariant set to f would be the empty set and If L::.01 = 2m - 1. Such
designs exist. For example, the Paley designs defined in [11, p97] are examples of
Hadamard designs with transitive automorphism groups and they exist whenever
4m - 1 is an odd prime power. Also the design of points and hyperplanes of the
d-dimensional projective space PG d(2) over GF (2) is an Hadamard design with
transitive automorphism group PGL (d + 1,2) and with m = 2d - l .
However, the bound in Theorem 3.1(a) is not best possible. Better bounds were
obtained in [8] for small values of m, namely for m ~ 31, combining the information
from Theorem 3.1(b) with the following technical result gave a better bound than
that in Theorem 3.I(a).
Theorem 3.4 [8, Theorem C] Let G be a group of permutations of a set 0 and
suppose that f ~ 0 has restricted movement with move (f) = m. Let t be the
number of G-orbits with which f has a proper nontrivial intersection. Then there
exists a G-invariant subset E of 0 such that

When t = 1 the inequality in Theorem 3.4 becomes IfL::.EI ~ 2m - 1, and

this bound can sometimes be attained as Example 3.3 showed. If f has restricted
202 Cheryl E. Praeger

movement m and the number t of G-orbits meeting I' in a non-empty, proper subset
is bounded (independently of m), then Theorem 3.4 gives an upper bound for II' .6.~1
which is linear in m. At the time of writing the first draft of this chapter, it was
not known whether this restriction on t was necessary in order to obtain a linear
upper bound on Ir.6.~I. In 1992, Peter Neumann, in a letter to Leonid Brailovsky,
asked the following question.
Question 2 (P. M. Neumann) Under the conditzons of Theorem 3.1(a), is it true
that if .6.~1 ::; 2m - 1 ?
In a letter to me, dated Match 19, 1995, Peter Neumann showed that Question 2
has an affirmative answer, so that we now have:
Theorelll 3.5 (P. M. Neumann) Let G be a group of permutations of a sct nand
suppose that l' ~ n has restricted movement with move (1') = m. Then there is a
G-invariant subset ~ oJn such that 11'.6.2.:1::; 2m - 1.

4 Infinite subsets with restricted movement

The results of Section 3 make it possible to prove a version of Theorem 2.1 for an
infinite subset with restricted movement. To see how this is done we review some
of the ideas behind the investigation in [8] and the proof of Theorem 3.1.
Recall from the discussion in Section :{ that a subset I' obtained from a G-
invariant subset by adding or removing a total of m points will have finite movement
at most m; in particular if 11'1 > m then r will have restricted movement (that is,
move (I') < 11'1) and subsets I' obtained in this way are called generic subsets with
restricted movement at most m.
Suppose that G is a group of permutations of a set 0. and that r is a (not
necessarily finite) subset of n with restricted movement. Let m := move(r), so
m is finite and If! > rrl. Observe that, if some G-orbit 0.' in n is contained in
1', then from the definition of restricted movement it is clear that l' \ 0.' also has
finite movement equal to m. Suppose that we remove from r all G-orbits which are
completely contained in 1'. The resulting set 1" has movement m, and either r' has
restricted movement or if'l = m. If 11"1 = m, then the original set I' was a generic
subset with restricted movement m.
So we shall suppose that ifl > m, move(r) = m, and for all G-orbits n', I'nn'
is a proper subset of n'. Now if r n n' = 0 for some n' then r still has movement
m under the permutation group induced by G on n \ n'. So we may assume that
r n n' f. 0 for all G-orbits 0.'.
Suppose next that, for some G-orbit 0.', 1" := I' n n' is infinite. It follows
from the definition of movement that move (r') ::; m, and since r' is infinite, that
1" has restricted movement. Then by a result of Peter Neumann [8, Theorem 2.4],
n' \ 1" is finite (in fact of size at most rn by [8, Corollary 2.5]). It follows from [8,
Lemma 3.1] that the set (r un') \ r', obtained from I' by replacing 1" with 0.' \ r',
The Separation Theorem for Group Actions 203

has movement equal to m. Thus, by making such a substitution for all infinite orbit
intersections of r (if any) we obtain a subset r" with movement m which intersects
each G-orbit 0' in a finite, non-empty, proper subset of 0'. Again, either r" has
restricted movement or r" has size m, and in the latter case the original set r must
have been a generic subset with restricted movement m.
So we now suppose that If! > m, move (r) = m, and for all G-orbits 0',
r' r n 0' is finite, is non-empty, and is a proper subset of 0'. In a similar
:=
manner, if for some G-orbit 0' the cardinality of r' is greater than that of 0' \ r',
then by [8, Lemma 3.1] the set (r U 0') \ r' obtained from r by replacing r' with
0' \ r' has movement equal to m. By making such a substitution for all orbit
intersections r nO' for which Ir n 0'1 > 10' \ rl (if any) we obtain a subset r"
with movement m which intersects each G-orbit 0' in a finite, non-empty, proper
subset of cardinality at most 10' \ r" I. Again, either r" has restricted movement
or r" has size m, and in the latter case the original set r must have been a generic
subset with restricted movement m. Moreover, by [3], all subsets r with movement
m = 1 < Irl are generic.
Finally, we assume that m 2 2 and that r" has size greater than m, so that
r" has restricted movement. It follows from Theorem 3.5 that 1f"1 :s: 2m - 1. Set
k := If''l, so m < k :s: 2m - 1. Applying Theorem 2.4 to r" we find that the
smallest G-orbit intersecting r" nontrivially (and hence intersecting r nontrivially)
has length at most (k2 - m)j(k -m). Now using elementary calculus we see that the
function f(x) := (x 2 -m)j(x-m) is strictly decreasing on the interval [m+ 1, 2m-I],
and so the maximum value of f(x) on this interval is f(m + 1) = m 2 + m + 1. On
the other hand, if k is known to be significantly larger than m, say k 2 ko, then
f(k) will be bounded above by f(k o ). For example, if ko = a . m > m + 1 then

Thus we have proved the following result (which is essentially [30, Theorem 2]
updated by the improved bound of Theorem 3.5).

Theorem 4.1 Let G be a group of permutations of a set 0 and suppose that

r ~ 0 (a not necessarily finite subset) has restricted movement with move (r) = m.
Then either
(a) r is a generic subset with restricted movement Tn (that is r is obtained from a
G-invariant subset by adding 01' removing a total of m points) 01'
(b) the smallest G-orbli intersecting r nontrivially has length at most m 2 + m + 1.
Moreover m 2 2, and ifr differs from a G-invariant set by more than a· m points,
where a 2 (m + l)jm, then the smallest G-orbit meeting r nontrivially has length
less than a 2 mj(a - 1).
204 Cheryl E. Praeger

5 Separation under intransitive group actions

Suppose now that G is a permutation group on a set 0 which is intransitive on OJ

and that r ~ 0 has restricted movement with move (r) = m. If r were contained in
a G-orbit 0' then we could consider the transitive action induced by G on 0', and
in this case the results discussed in Section 2 give the best possible upper bounds
on the length of 0'. Thus in this section we shall be interested in the case where r
intersects nontrivially at least two of the G-orbits in O. Suppose that we can write
o = El U E2 where Ei is G-invariant, and r; := r n Ei :/= 0, for i = 1,2. It is
certainly the case that move (r i) ::; m and even a little more is true.

Theorem 5.1 [8, Proposition 4.1] Let G be a group of permutations of a set 0

and suppose that r ~ 0 has restricted movement with move (r) = m. Suppose
that 0 is the disjoint union of G-invariant sets El and E2 such that, for i = 1,2,
ri := Ei n r :/= 0. Then

move(rd + move (r2) ::; 3m/2.

First we note that the bound in this result can be attained for some values of
m. Let m = 2r ~ 2 and let G, 0 and r be as in Example 3.2. Choose a non-identity
element 9 E G, let El be the union of the G-orbits Oi on which 9 acts nontrivially,
and let E2 := 0 \ E 1 . Then move(r) = move (r l ) = m and move (r 2) = m/2.
However, even though move (ri) ::; move (r), with strict inequality for at least
one i, it is possible for neither of r 1, r 2 to have restricted movement (see Exam-
ple 5.2 below). This illustrates the fact that results about separation of subsets
by intransitive groups may not be deduced immediately from similar results for
transitive groups. The intransitive case must be considered separately.
Example 5.2 Let m = k - 1 be a prime power, let 0 be the set of points of
the Desarguesian projective plane PG 2(m), let El and r be the sets of points on
distinct lines of the plane, and let G be the setwise stabiliser of El in PGL (3, m).
Then the G-orbits in 0 are EI and E2 := 0\E 1 , and we have IEII = k = m+ 1 and
IE21 = (k _1)2 = m 2. Moreover r is a k-element subset with move(r) = m < k,
and setting r i := f n Ei for i = 1,2, we have 1ft! = move(f 1 ) = 1 and
1f21 = move (f2) = m.
Before proceeding with our discussion we note that the proof of Theorem 3.4 is
a straightforward induction on the number t of orbits meeting the set r; the proof
uses Theorem 3.1(c) to start the induction for t = 1, and uses Theorem 5.1 for the
inductive step.
If f is a finite set with restricted movement intersecting exactly t (1 ::; t ::; Ifl)
of the G-orbits nontrivially, then the best general separation result available [17,
 The Separation Theorem for Group Actions 205

Theorem 1.4] (or see [29, Corollary 1.5] when t = 2) is a generalisation of Theo-
rem 2.1 which gives an upper bound for the length of the smallest G-orbit in terms
of both If! and t.
Theorem 5.3 [17, Theorem 1.4] Let G be an intransitive permutation group on a
set 11, and let r be a k-element subset of11 with restricted movement. Ifr intersects
nontrivially t of the G-orbits in 11, where 1 ~ t ~ k, then the smallest of these orbits
has length at most (k - t + 1)2 + t - 2.

When t = k and t -1 is a prime power, then this bound can always be attained.
= =
Example 5.4 Let k t q + 1 for some prime power q, let 11 be the set of lines
of the Desarguesian affine plane AG 2(q), and let G be the group of translations of
AG 2(q). Then 1111 = q(q + 1), G is elementary abelian of order q2, and G has q + 1
,orbits of length q in 11, namely the q+ 1 parallel classes oflines. Moreover, each non-
identity translation in G fixes each line in exactly one of the parallel classes and acts
semiregularly on the lines in each of the other parallel classes. Let r C 11 consist
of exactly one line from each parallel class. Then Irl = q+ 1 = k and, from our
observation about the action of each non-identity translation g on 11, Ir9 n f! = 1,
whence move(r) = q < k. Thus r has restricted movement, r meets t = k orbits,
and each orbit has size q = (k - t + 1)2 + t - 2.

In fact, the examples given above are essentially the only kinds of examples for
which t = k 2: 3 and the bound of Theorem 5.3 is attained.

Theorem 5.5 [17, Theorem 1.5] Let G be an intransitive permutation group on a

set 11, and let r be a k-element subset of 11 with restricted movement, where k 2: 3.
Suppose further that r intersects nontrivially k of the G-orbits in 11, and that the
smallest of these orbits has length k - 1 (the bound of Theorem 5.3). Then k - 1
is a prime power, and, if 11' is the union of the G-orbits meeting r, then 11' can be
identified with the set of lines of an affine translation plane of order k - 1 in such a
way that r consists of one line from each parallel class, GO' fixes each parallel class
setwise, and GO' contains the group of translations.

Although the case t = k 2: 3 is the only one where (t 2: 2 and) we have a

classification of the extreme examples, we do have some fairly restrictive information
about the case where t = 2 and the bound of Theorem 5.3 is attained.

Theorem 5.6 [29, Theorem 1.4] Let G be an intransitive permutation group on a

set 11, and let r be a k-element subset ofo. with restricted movement which intersects
nontrivially 2 of the G-orbits, 0. 1 and 0. 2. Then either the smaller of 0. 1 and 112
has length less than (k _1)2, or Ir n 0. 11 say is equal to 1 and Ir n 0. 21= k -1, and
either
(a) 10.21= (k - 1)2 and 11111 ~ IGI/(k - 1), or
(b) 11111 = =
(k _1)2, 11121 (k _1)2 + I, and IGI2: 10.11·111 21(k - 1).
206 Cheryl E. Praeger

The conditions of parts (a) and (b) hold when k = 2, G is a cyclic group of
order 2 acting intransitively with two orbits of lengths 1 and 2, and r consists of
one point from each G-orbit.
Problem 1 Determine all values of k for which the conditions of Theorem 5.6(b)
can hold.
It is not known whether the bound of Theorem 5.3 can be attained for other
values of t and k.
Problem 2 Determine all values of t and k for which the upper bound for the
minimal orbit length in Theorem 5.3 can be attained, and classify those groups G
and subsets r attaining the bound.
As we remarked in Section 2, if r is a k-element subset of n with restricted
movement under the action of a group G, then the collection E := {rglg E G} is the
set of edges of an intersecting k-uniform hypergraph with vertex set n. The work of
Fiiredi and others on fractional transversals of such hypergraphs is relevant to some
of the questions discussed in this chapter. A fractional transversal of the hypergraph
(n, E) is a non-negative real-valued function tr on n such that L-YEr g tr(i) 2: 1 for
all r g E E. Define Itrl := LaEiI tr(a). Then the fractional transversal number T*
for (n, E) is defined to be the minimum of Itrl over all fractional transversals tr. Let
n 1 , ... , n t be the G-orbits, and set k i := In i n rl. By averaging over the G-orbits
we may assume that an optimal fractional transversal tr on (n, E) is constant on
each of the ni, say tr takes the value ai on n i . Then we have T* = L1<i<tailnil,
:s :s
and L 1 i taiki = l. If the minimum value of In i II k i occurs for i- ~ j, then
these equations imply that

Thus upper bounds on T* will yield information about the minimum length of the
G-orbits. The inequality above together with the upper bound k - 1 + k- 1 on T*
(see [14, Corollary 4]) yields the upper bound of Theorem 2.3 on the minimal length
of the G-orbits. For t 2: 2 however, the best known bounds on T* yield a slightly
weaker upper bound on the minimal orbit length than that given in Theorem 5.3.
The paper [15] of Fiiredi contains a survey of some of the relevant results about
hypergraphs, and it has a good bibliography.
We end this section with a general discussion of the relation between the orbit
length bounds which can be obtained and the distribution of the points of the set
r amongst the G-orbits.
Suppose that r is a k-element subset with restricted movement and that r
meets exactly t of the G-orbits, 0 1 , ... , Ot. Set ri := r n n i and Vi := Ird for each
i. Then each Vi # 0, LVi = k, and we may renumber the Oi so that V1 :s ... :sVt
and hence Vt 2: kit. In the proof of Theorem 5.3 an upper bound on the length
of the smallest of the Oi is obtained as an expression involving V1, ... , Vt. This
The Separation Theorem for Group Actions 207

expression is greatest when (V1, ... , Vt) = (1, ... , 1, k - t + 1), thus yielding the
upper bound in Theorem 5.3.
We give an illustration of how the bound may be improved when the distribu-
tion of the points of f over the Oi is not as unbalanced as it is in the configuration
corresponding to the upper bound of Theorem 5.3. Consider the requirement that
Vt ~ k - xk/t for some positive real number x. Since Vt ~ kit we must have
x ~ t - 1. Also, since Vt = k - L:i<t Vi ~ k - t + 1, this inequality represents a
genuine restriction on Vt only if k - xk/t ~ k - t + 1, that is, x ~ t(t - 1)/k. Thus
we are interested only in values of x satisfying t(t - l)/k ~ x ~ t - 1.
Theorem 5.7 [17, Theorem 1.5] Suppose that G is an intransitive permutation
group on a finite set 0 with orbits 0 1 , ... ,Ot, and that f is a k-element subset of
o with restricted movement and f meets Oi in Vi points for i = 1, ... , t, where
o < V1 ~ ... ~ Vt. If x is a real number satisfying t(t - l)/k ~ x ~ t - 1, and if
Vt ~ k - xk/t, then the smallest of the Oi has length at most

e(1- 2X(:; x)) _ (t _ l)(t - 2).

To see that this can be quite powerful in certain situations, consider the case
where f has restricted movement, k =
If I ~ 4, and f meets t =
3 orbits 0 1 , O 2 , 0 3 ;
taking x = 3/2, Theorem 5.7 says that the smallest of the three Oi has length at
most k 2 /2 - 2 provided that each f n 0i contains no more than half the points of
f. This is much stronger than the bound from Theorem 5.3 which in this case is
(k-2)2+1.

6 Permutation groups with bounded movement

This section is devoted to a discussion of permutation groups G on a set 0 for which

many subsets of 0 have finite movement at most m, for some given positive integer
m. In particular, if move (f) ~ m for all f ~ 0, then G is said to have bounded
movement and the movement of G is defined as the maximum of move (f) over all
subsets f. Having bounded movement is a very strong restriction on a group, but it
is natural to ask just which permutation groups have bounded movement m. The
question was first posed by Carlo Casolo during a discussion we had in 1989. The
bounded movement condition was a generalisation to arbitrary group actions of a
condition he was working with for the action of a finite soluble group on a module
over a finite field.
Suppose now that G is a permutation group on a set 0 and that G has bounded
movement equal to m. Quite a lot can be said about G simply by applying results
from the previous sections. For example, the group of permutations induced by G on
one of its orbits Of also has movement at most m, and it follows from Theorem 2.1,
208 Cheryl E. Praeger

on considering subsets of 0' of size at most m + 1, that 10'1 :s: (m + 1)2. Hence all
G-orbits are finite of length at most (m + 1)2 (and of course better bounds, which
are still quadratic in m, follow from Theorems 2.3 and 2.4). Also, by considering a
subset of points consisting of one point from each nontrivial G-orbit (that is each
orbit of size greater than 1), it follows from Theorem 3.1(b) that G has less than
2m nontrivial orbits in O. Thus the number of points of 0 moved by G is less than
2m(m + 1)2. An investigation in [28] of groups with bounded movement, using the
Separation Theorem and some careful but elementary arguments, resulted in the
following theorem.
Theorem 6.1 [28, Theorem 1] Let G be a group of permutations of a set 0 with
no fixed points in 0, and suppose that, for some positive integer m, G has bounded
movement equal to m. Then
(a) the number t of G-orbits is at most 2m - 1,
(b) each G-orbit has length at most 3m, and moreover
(c) 101:s: 3m + t - 1 :s: 5m - 2.
The bounds in Theorem 6.1( a) and (b) can be attained for certain values of m,
and therefore the bound in part (c) can also be attained for certain m when t = 1.
First we consider the bound of part (a). The family of permutation groups
defined in Example 3.2 consists of permutation groups G = Z;+1 with 2r +1 - 1
orbits of length 2 and with the property that each nontrivial element of G has 2r
cycles of length 2. Thus, for any subset r of points and any 9 E G, the set ru \ r
consists of at most one point from each of the G-orbits on which 9 acts nontrivially,
and consequently lru \ rl :s: 2r. It follows that G has bounded movement m = 2r
and G has 2m - 1 nontrivial orbits. This family of examples is essentially the only
one to attain the bound.
Theorem 6.2 [10] Let G be a group of permutations of a set 0 with no fixed points
in 0, and suppose that, for some positive integer m, G has bounded movement equal
to m and has 2m - 1 orbits. Then G is an elementary abelian 2-group, all the G-
orbits have length 2, and m = 2r for some non-negative integer r.
Let G be a group satisfying the conditions of Theorem 6.2, and let the G-orbits
in 0 be 0 1 , ... , 02m-1, where m = 2r. Then G is a subgroup of the direct product
G 1 x ... X G 2m - 1 where Gi == Z2 is the permutation group induced by G on Oi
for each i. Further, from our discussion before the statement of Theorem 6.2, each
element of G must act nontrivially on at most m of the Oi. However it is not known
precisely which sub direct products of G 1 x ... X G 2m - 1 have movement equal to m.
Question 3 What are the possible orders of the groups G in Theorem 6.2?
Now we turn to the bound in part (b) of Theorem 6.1. It was shown in [28,
Theorem 2] that, for a transitive group G on a set 0 of size 3m to have movement
m, m must have the form 2a 3b 5c with a and c at most 1. Moreover [28, Theorem 2]
also gave several examples of such groups, namely transitive groups of exponent
The Separation Theorem for Group Actions 209

3 on a set of size 3m = 3r ~ 3, the symmetric group S3 of degree 3m = 3, and

the groups A4 and A5 acting transitively of degree 3m = 6. It has recently been
proved that these are the only examples. In [16] it was conjectured that the only
examples were S3, A 4 , A5 and transitive 3-groups, and it was shown that a minimal
counterexample must be a nonabelian simple group and must be primitive on O.
Then in [23] it was shown that every example which is a 3-group must have exponent
3, and that the only simple primitive example is A5 of degree 6.
Theorem 6.3 [16, 19] Let G be a transitive group of permutations of a set 0 of
size 3m for some positive integer m. Then G has bounded movement equal to m if
and only if m = 3r ~ 1 and G has exponent 3, or m = 1 and G = S3, or m = 2
and G = A4 or A 5.
The results in Theorems 6.1 amd 6.3 highlight the special role of the prime 3 in
this study. For groups with order relatively prime to 3 we obtain different bounds.
Two lemmas from [28] give a starting point.
Theorem 6.4 [28, Lemmas 2.1 and 2.2] Let G be a group of permutations of a
set 0 with no fixed points in 0, and suppose that, for some positive integers m and
t, G has bounded movement equal to m and G has t orbits in O.
(aJ If G zs a 2-group then each G-orbit has length at most 2m
and 101 :::; t + 2m - 1 :::; 4m - 2.
(b J If G is not a 2-group and if p is the least odd prime dividing IGI, then each
G-orbit has length less than 1+2mp/(p-l) and 101 < t+2mp/(p-l). In particular,
if p ~ 5, then 101 :::; (9m - 3)/2.
It is easy to show that every permutation group on a set of size at most 2m has
bounded movement at most m (see [28, Lemma 3.4]). So the groups of interest (in
the context of Theorem 6.4) are those which are not 2-groups and which have order
relatively prime to 3. In particular it would be interesting to have an analogue of
Theorem 6.3 in this case.
Problem 3 Let m be a positive integer and p an odd prime. Determine all tran-
sitive permutation groups G on a set of size i2mp/(p - 1)1 which have bounded
movement equal to m, and for which p is the least odd prime divisor of IGI.
There certainly are some families of examples as the next lemma shows.
Lemma 6.5 Let p be an odd prime and let m := pa-l(p - 1)/2 for some a ~ 1.
(aJ A transitive permutation group of exponent p on a set 0 of size pa = 2mp/(p-l)
has bounded movement equal to m.
(bJ Lei a= 1 and let s= 2b be the highest power of2 dividing p-l. Let Fsp denote
the Frobenius group of order sp acting transitively on a set 0 of size p = 2mp/(p-l).
Then each transitzve subgroup G of Fsp has bounded movement equal to m.
Proof. (a) Let g E G \ {I} and let r ~ O. For each cycle C of g of length p,
(r n C)9 \ (r n C) has size at most (p - 1)/2. Thus Ir 9 \ rl is at most the number
210 Cheryl E. Praeger

of cycles of 9 of length p times (p - 1)/2, that is, \r 9 \ r\ ::; (\O\/p)· (p - 1)/2 = m.

Thus G has movement at most m. Moreover, since G is transitive, the average
number of fixed points of elements of G is 1 and so G contains an element g, say,
with no fixed points in 0; for this element g, a subset r consisting of "every second
element of every cycle of g" (that is, if a cycle C of 9 is C = (0:1, ... , O:p) then put
into r the (p - 1)/2 points 0:2,0:4, ... , O:p-d has size m and is mapped disjointly
from itself by g. Thus the movement of G is equal to m.
(b) By part (a), if 9 E G has order p, then lru \ rl ::; m = (p - 1)/2 for all
subsets r and there is a subset r such that r g \ r has size equal to m. Suppose
now that 9 E G \ {l} has order o(g) a power 2. Then 9 has one fixed point and
(p - l)/o(g) cycles of length o(g) in O. For each r ~ 0, ru \ r consists of at most
0(g)/2 points from each cycle of 9 of length o(g), and hence has size at most m.
Thus G has bounded movement equal to m.
Work on Problem 3 is currently being carried out by Mehdi Khayarty and
Sjabon Sedghi, two PhD students of Akbar Hassani in Tehran. Further, Akbar
Hassani is developing a full version of Theorem 6.1 for groups G such that the
smallest prime divisor p of IGI is at least 5.
It is not clear whether the bound in part (c) of Theorem 6.1 can be attained
for large values of t. Put in another way we could ask:
Question 4 For which positive integers n is there a positive integer t and a per-
mutation group G on a set of size n with t nontrivial orbits such that G has bounded
movement equal to I(n - t + 1)/31 ?
The paper [28, Lemma 3.6) contains a construction of such a group if 3 divides
n and the 3-adic expansion of n (namely n = Li>lO a j 3i with each aj equal to
(J,-I, or 2) satisfies 1 ::; Li> 1 ai ::; 3. The number t of nontrivial orbits in the
construction is the number ot terms of the 3-adic expansion, that is, t = Li> 1 ai,
and so the greatest value of t for this family is 3. It is not known whether the bound
of Theorem 6.1 ( c) is best possible for larger values of t.
Recently these results about permutation groups G on 0 with bounded move-
ment were generalised. The assumption of restricted movement for essentially all
subsets of 0 was weakened to the requirement that, for some integers m and k with
o < m < k, all k-element subsets of 0 should have movement at most m. If the
size of 0 is at most k + m then it follows from the definition of movement that this
condition is satisfied for every permutation group G on O. Thus it is the case where
101 > k + m which is of interest. Surprisingly, in this case, it turns out that the
group G must have bounded movement at most m, and so Theorem 6.1 applies.
Theorem 6.6 [30, Theorem 3) Let G be a group of permutations of a set 0 with
no fixed points zn 0, and let m and k be posztive zntegers such that m < k and
101 > k + m. Suppose further that move (r) ::; m for all k-element subsets r of O.
Then move(r) ::; m for every subset r of 0 (and hence the number t of G-orbits
is at most 2m - 1, Inl ::; 3m + t - 1 ::; 5m - 2, and k ::; 2m + t - 2 ::; 4m - 3).
 The Separation Theorem for Group Actions 211

Moreover, if some G-orbit n' has length greater than k + m, then k ::; 2m - 1 and
In'l ::; 3m.
Observe that Theorem 2.5 or Theorem 6.6 gives restrictions on the lengths of
the smallest G-orbit or of all G-orbits, when we restrict the movement of one or
all k-element subsets of points, respectively. It would be interesting to know what
information can be deduced about the G-orbits when we restrict the average of the
movements of k-element subsets.
Question 5 What information about the orbits of a permutation group G on a
finite set n can be deduced from restricting the average of the movements move (f)
of k-element subsets r to be at most m, for some m < k?

7 Groups with a small squaring property

A group G is said to satisfy the small squaring property on k-sets (where k is

an integer, k ~ 2) denoted SS (k) if, for each k-element subset K of G, the set
K2 which is defined as {ab I a, b E K} has size strictly less than k 2. Clearly
any abelian group G will have the the small squaring property on k-sets for every
k ~ 2, and the properties SS (k) for various k may be viewed as generalisations of
the property of being abelian. The first investigation of groups with one of these
properties was carried out by G. A. Freiman. In 1981 he classified groups with the
property 88 (2) (showing that they were precisely abelian groups and Hamiltonian
2-groups), and began a study of groups with the property SS (3). A significant part
of the classification of groups with the property SS (3) was carried out by Ja. G.
Berkovich, Freiman and the author in [1], and the classification was completed by
Patrizia Longobardi and Mercede Maj in [22].
Theorem 7.1 [1, 13] and [22, Theorems A and C) (aj A group G has the small
squaring property 8S (2) if and only if either G is abe/ian or G =
E X Q where E is
an elementary abelian 2-group and Q is the quaternion group of order 8.
(b) A group G has the small squaring property S8 (3) if and only if one of the
following holds.
(i) G is abelian;
(ii) G = A(x) where A is abelian, x 2 E A, and aX = a-I for all a E A;
(iii) G is a nonabelian 2-group of exponent 4, and either (x 2 I x E G) has order
2, or G = E x D, where E is an elementary abelian 2-group and D is one of two
explicitly given groups, one of order 32 and the other of order 64.
A pattern was emerging here: the examples tended to have either a very large
abelian normal subgroup or a very large abelian quotient group. The significance
of this pattern would become more evident later. In 1989 Peter Neumann showed
that a group having the small squaring property 88 (k) for some k ~ 2 must be
finite-by-abelian-by-finite. His proof involved an ingenious and unexpected use of
the Separation Theorem.
212 Cheryl E. Praeger

Theorem 7.2 [20, Theorem 1] and [26] If a group G has the small squaring
property SS (k) for some k ~ 2, then G has normal subgroups Hand N, with
N ~ H, such that H / N is abelian and both G / Hand N are finite.
It was the quantitative version Theorem 2.1 of the Separation Theorem which
was used in the proof of this result to construct the normal subgroup H. We will
review the ideas of this part of the proof
Suppose that a group G has the property SS (k) where k is some integer, k ~ 2,
and let S be the set of all elements 9 E G such that the number of conjugates of 9 by
elements of G is at most k 2 (k _1)2; so S is the union of all "small" conjugacy classes
of G. The idea is to consider the conjugation action of G on itself. Under this action
S is a G-invariant subset and we may consider the action of G by conjugation on
n := G \ S. All G-orbits in n have length strictly greater than k2 (k - 1)2 (by the
definition of S), and consequently by Theorem 21 any two subsets of n of size at
most k(k - 1) may be separated completely from each other by some element of G.
This fact was used as follows to prove that G can be written as the union of
some collection of k - 1 translates of S. Suppose that this is not the case. Then
k distinct elements Yl, ... , Yk of G can be found such that Yi Yj 1 ¢ S for all i < j.
Then also, since yiyjl is conjugate to yj1Yi and since S is closed under taking
inverses, Y; 1 Yj ¢ S for all i < j. It follows that, for all i ::j:. j, both YiYj 1 and
y;l Yj are elements of n. Set r := { yiYj 1 I i ::j:. j} and 6. := {y;lYj I i ::j:. j}, so
that rand 6. are both subsets of n of size at most k(k - 1). By Theorem 2.1,
=
there exists an element 9 E G such that r n 6. g 0. This means that, for all i ::j:. j
and i' ::j:. j', YiYj 1 ::j:. g-l Yi-; 1Yjl g, that is, (Yi9 )(Yi 9 )-1 ::j:. (Yi' g) -1 (Yjl g), which in
turn is equivalent to (YiI9)(Yig) ::j:. (Yjlg)(Yjg). The last inequality, for all i::j:. j and
i' ::j:. j', implies that the k-element subset K := {Y1g, ... ,Ykg} satisfies IK21 = k 2 ,
contradicting the fact that G has the small squaring property SS (k).
Thus there are elements Y1, ... ,Yk-1 E G such that G = Ui> 1 SYi. Let H be
the subgroup of G generated by S. Since S is closed under conjugation, H is a
normal subgroup of G, and since SYi ~ HYi for all i, H has index at most k - 1
in G. The proof of Theorem 7.2 is completed (in [20] or [26]) by showing that the
commutator subgroup N := [H, H] has order bounded above by some function of
k.
The information given by Theorem 7.2 was refined in [20] by Herzog, Longo-
bardi and Maj to show that examples of groups with the small squaring property
SS (k), for some k 2: 2, may be roughly divided into two classes, one with a large
abelian normal subgroup, and the other with a large abelian quotient group, a
division suggested by the classification of groups with the property SS (3) in The-
orem 7 .1(b). The statement requires the following definitions. A group G is said
to be nearly dihedral if it contains a normal abelian subgroup H of finite index, on
which each element of G acts by conjugation either as the identity automorphism
or as the inverting automorphism. The subgroup of a group G generated by the
squares of all elements of G will be denoted by G(2).
 The Separation Theorem for Group Actions 213

Theorem 7.3 [20, Theorem 2] A group G has the small squaring property SS (k),
for some k 2: 2, if and only if either G is nearly dihedral or G(2) is of finite order.

Now we finish one part of the story of the investigation of groups with a small
squaring property. Since the small squaring property SS (k), for some k 2: 2, forces
a group to be "nearly abelian" in the sense of Theorem 7.2 or Theorem 7.3, is
there a somewhat stronger condition of the same nature which forces a group to
be abelian? More precisely, if G is abelian then, for all k-element subsets J{ of
G, IJ{21 ~ k(k + 1)/2, and Peter Neumann (see [6]) asked whether the condition
IJ{21 ~ ak 2 for every k-element subset of G (where k 2: 2 and a is a fixed real
°
number satisfying < a < 1) might imply that G is abelian.
Brailovsky [4] first showed that, if IJ{21 ~ k(k + 1)/2 for all k-element subsets
J{ of G, then G is abelian (where k > 2 and if G is finite then IGI > 2k(k 2 - 1)).
Next he improved this by showing in [5] that, if IJ{21 ~ k(k + 1)/2 + (k - 3)/2 for
all k-element subsets J{ of G, then G is abelian (where k > 2 and if G is finite
then IGI > 15(k - 1)(k 2 - 3». These results led him to define a squaring bound for
infinite abelian groups to be a function f(k) of the natural numbers such that, if
there exists an integer k 2: 2 such that IJ{21 ~ f(k) for all k-element subsets J{ of
G, then G is abelian. He sought the best possible squaring bound for infinite abelian
groups, by which he meant a squaring bound f(k) for infinite abelian groups such
that the following condition holds:
If g(k) is a function on the natural numbers such that g(k) > f(k) for some
natural number k, then g(k) is not a squaring bound for infinite abelian groups.
He was able to construct such a best possible bound [6, Theorem], namely

r
One consequence of this result is that f( k) = ak21 is a squaring bound for infinite
°
abelian groups if and only if < a < 5/6. The reader may be interested to know
that the proof of the result of [5] makes essential use of the result in [3], which was
the principal motivation for the investigation described in Section 3.

8 A probabilistic version of the Separation Theorem

For some purposes an understanding of the Separation Theorem from a probabilistic

point of view is helpful. Suppose that G is a permutation group on a finite set n
and that r, d are subsets of n of size k and n respectively. Consider an experiment
in which we randomly select an element 9 from G in such a way that the probability
of selecting any given element of Gis l/IGI; we construct the image P of r under
9 and find the size X(g) of the intersection r g n d. Then X := X(g) is a random
variable which takes integer values (which of course depend on the element g chosen)
in the closed interval 1:= [O,min{k,n}].
214 Cheryl E. Praeger

We determine the expected value E(X) and the variance var(X) of X. In

general E(X) depends on the distribution of the points of I' and ~ over the G-
orbits, but if G is transitive on n, then E(X), often written X, depends only on
k, n and the size of n. The expression for the variance var (X) (that is, the expected
value of (X - X)2) is more complicated than that for E(X) and it depends on the
distribution of the ordered pairs of points of I' and of ~ among the G-orbits on
ordered pairs of points from n. When there are few G-orbits on ordered pairs from
n, the expression for var (X) is of course simpler, and we give it explicitly, as a
function of k, nand Inl, in the case where G is 2-transitive on n. (Recall that G
induces a natural action on n x n by (o:,(3)g := (o:g,fJU) for 0:,(3 En and 9 E G.
Let ~(1), ... , ~(r) denote the G-orbits in n x n. The expression for var (X) involves
the number of ordered pairs from I' and ~ which lie in ~(j), for each j.)

Theorelll 8.1 [17, Section 7] Lei G be a group of permutations of a finite set n,

and let 1', ~ be subsets of n of size k and n respectively. Let X be a random variable
defined by X := X(g) = lf9n~L where 9 is a uniformly distributed random element
ofG.
(a) If nj , j E J, is the set of G-orbits in n, then the expected value of X IS

In particular when G is transitive, E(X) = kn/lnl.

(b)L et "<'(I)
u , ... , "<'(r)b
u h G- or'b'ds zn
etc ' .H .X.H,
. an d, jar
f J'-1
- . I
, ... 1, et '
YJ d
an '
zJ
denote the number of pairs (0: ,(3) E ~(j) such that {o:, (3} ~ I' and {o:, (3} ~ ~
respectively. Then the variance of X is

var (X) = '"

L Yj' Zj
/~(j)/ - /.L'(X)2
'" .
l:Sj:Sr

In particular, if G is 2-transitive on n, then

(X) _ k n (Inl- k) (In/- n)

var - In/2 (Inl- 1) .

It is a simple matter to deduce Theorem 2.1 from Theorem 8.1(a), and the
arguments used in the proof of Theorem 8.1 (a) are essentially those in the proof of
Theorem 2.1: if all the G-orbits have length greater than kn then E(X) is strictly
less than (2: jE J II' n njl'l~ n njl)/kn by Theorem 8.1, and this expression is at
most (2: jO II' n nj I) . (2: jO 16. n nj I)/kn = 1. It follows that X(g) must be zero
for some 9 E (;, that is f9 n ~ = 0 for some 9 E G.
The Separation Theorem for Group Actions 215

9 Combinatorial applications of the Separation Theorem

The version of the Separation Theorem stated as Theorem 2.1 was proved inde-
pendently in 1977 by B. Ganter, J. Pelikan and L. Tierlinck in the situation where
the set n is the collection (~) of all k-element subsets of a v-element set S, and
the group G is Sym (S) in its natural action on n. Their motivation came from
combinatorial considerations. If k = 2 then any subset r of n may be identified
with (the edge set of) a graph with vertex set S; for larger values of k, a subset r
of n satisfying certain extra conditions may be identified with (the block set of) a
combinatorial design.
Throughout this section we assume, therefore, that n = (n =
and G SymeS)
for some set S of size v and some integer k, with 2 ~ k < v.
In [18] a subset r of n with restricted movement is said to be sprawling. The
main aim of the paper [18] was to analyse the behaviour of the function b(k, v)
of the integer variables k and v (0 < k < v) where b(k, v) is defined as the least
positive integer n such that there exists a sprawling n-element subset r ~ (V, for
lSI = v. Clearly b(k,v) = b(v - k,v) for all k and v, and b(l,v) = Lv/2J + 1 for
all v. In [18, Lemma 1) it was shown that b(k, v) ;::: m 1 / 2 for all k, v. This is

precisely the information given by the last sentence of Theorem 2.1 in the case of
= =
G Sym (S) and n (f). The remaining assertions of Theorem 2.1 were obtained
for this group action in [18, Lemma 7]. Asymptotic results about the growth of
b(k, v) were proved in [18, Theorem 1] but, for example, the question of whether
or not b(k, v) is a monotonically increasing function in k (for k < v/2) or in v was
asked but not answered. In the special case where k = 2, so that a subset r of n
may be identified with a graph with vertex set S, it was shown in [18, Theorem 2]
that b(2, v) = v-I and all sprawling graphs with v vertices and v-I edges were
classified.
The problem of separating two subsets rand Ll of n = (f), for 2 < k < v,
arose again (see [19], and [31, Section 4.5]) as part of an investigation of construction
methods for combinatorial designs. The sets r and ~ were the block sets of two
t-(v, k,.\) designs with point sets S, and as the starting point for a method of
constructing new designs, it was necessary to produce computationally an image
f9 of r, for some 9 E SymeS), which was disjoint from Ll. The existence of such
an image r g is guaranteed by Theorem 2.1 if Irl·I6.1 < (~), that is, if v is "large
enough" with respect to k. However Theorem 2.1 takes no account of any special
structure or special properties of the sets rand 6., and G. B. Khosrovshahi has
conjectured that such an image exists for many other nontrivial t-(v, k,.\) designs.
Recall that, for integers 2 ~ t < k < v and .\ > 0, a t-(v, k,.\) design consists
of a set S of v points and a family of k-element subsets of S, called blocks, such
that each t-element subset of S is contained in precisely .\ blocks. The trivial design
based on S is the one in which every k-element subset of S is a block, and all other
t-designs are said to be nontrivial. Since the set of complements of the blocks of a
216 Cheryl E. Praeger

t-( v, k, A) design forms the block set of a t-( v, v - k, A') design, for some ,AI, we shall
assume from now on that k .:; v/2.
Conjecture (Khosrovshahi) Suppose that t, k, v, A are positive integers
(2 .:; t < k .:; v/2) such that a nontrivial t-(v, k, A) design exists, and such that
no t -( v, k,,AI) design exists for any ,AI < A. [f r is the block set of such a design,
then another t-(v, k, A) design exists, with block set Ll say, such that r n Ll = 0.
A strong version of t.his conject.ure is true for 2-( v, k, 1) designs, namely, for
r, Ll the block sets of two 2-(v, k, 1) designs on the same point set 5, there exists
an element 9 E Sym (5) such that ru n Ll = 0. An algorithm for finding an image
ru disjoint from Ll is given in [31, Section 4.5].
Suppose that r, Ll are the block sets of two t-( v, k, A) designs with the same
point. set 5. While not. providing proof of the existence of an image r 9 disjoint from
Ll, or giving an estimate of the likelihood of finding such an image computationally,
some insight into the distribution of the intersection sizes X(g) := Ir 9 n LlI, for
9 E Sym (5), can be gained from computing the expected value and the variance of
the random variable X := X (g) (where 9 is a uniformly distributed random element
of G). The tools for doing this were presented in t.he previous section. When applied
to this problem about combinatorial designs they yield the following results.
Theorem 9.1 [21] Let t, v, k, A be positive wtegers such that 2 .:; t < k .:; v/2.
Further' let G = Sym (5), where 5 is a set of size v, and consider the action of
G on the set n == (~). Let r, Ll be subsets of n which form the block sets of
two t-(v, k, A) designs with powt sets S. Let X be a random variable defined by
X := X(g) = Ir 9 n LlI, where 9 is a uniformly distributed mndom element of G.
Then the expected value of X is

A2 (V)
E(X) = ·k.
("k-t
_ t.) 2

Moreover, for j == 0, 1, ... , k, let Yj and Zj denote the number of pazrs (0:,;3) E n x n
with lG'n;31 = j such that {G', j3} ~ rand {G', j3} ~ Ll respectively. Then the variance
of X is
( V) = -(")
var./\.
1 ""
L...-k)Yj'(,,-k
Zj
- E();T)2
. .
k O-:;J-:;k Ck-)
The expression for the variance var (X) is quite complicated and depends in
general on the struct.ure of the design and not simply on the parameters t, v, k, A.
However, in the case of 2-( v, k, 1) designs, it turns out that var (X) depends only
on v and k.
Corollary 9.2 [21] Let v, k be posztzvc wtegers such that 2 < k':; v/2, let 5 be a
set of szze v, let G = Sym (5), lend consuicr the actwn of G on the set Sl = (~). f-et
The Separation Theorem for Group Actions 217

r,.6 be subsets of rl which form the block sets of two 2-( v, k, 1) designs with point
sets S. Let X be a random variable defined by X := X(g) = 1m n.61, where 9 is a
uniformly distributed random element of G. Then the expected value of X is

E(X)=~,
V-2)2(
k-2

and the variance of X is

where r = (v - l)j(k - 1).

References
[1] Ja. G. Berkovich, G. A. Freiman and C. E. Praeger, Small squaring and cubing
properties for finite groups, Bull. Austral. Math. Soc. 44(1991),429-450.
[2] B. J. Birch, R. G. Burns, S. O. Macdonald and P. M. Neumann, On the orbit
sizes of permutation groups containing elements separating finite subsets, Bull.
Austral. Math. Soc. 14(1976), 7-10.
[3] L. Brailovsky, Structure of quasi-invariant sets, Arch. Math. (Basel) 59(1992),
322-326.
[4] L. Brailovsky, A characterisation of abelian groups, Proc. Amer. Math. Soc.
117(1993),627-629.
[5] L. Brailovsky, On the small squaring and commutativity, Bull. London Math.
Soc. 25(1993), 330-336.
[6] L. Brailovsky, Combinatorial conditions forcing commutativity of an infinite
group, J. Algebra, 165(1994), 394 - 400.
[7] L. Brailovsky, D. V. Pasechnik and C. E. Praeger, Classification of 2-quasi-
invariant sets, Ars Combin., to appear.
[8] L. Brailovsky, D. V. Pasechnik and C. E. Praeger, Subsets close to invariant
subsets under group actions, Proc. Amer. Math. Soc., to appear.
[9] A. R. Calderbank, Symmetric designs as the solution of an extremal problem
in combinatorial set theory, European J. Combin. 8(1987), 171-173.
[10] J. R. Cho, P. S. Kim and C. E. Praeger, The maximum number of orbits of a
permutation group with bounded movement, (in preparation).
218 Cheryl E. Praeger

[11] P. Dembowski, Finite Geometries, Springer-Verlag, New York, (1967).

[12] P. Frankl and Z. Fiiredi, Finite projective spaces and intersecting hypergraphs,
Combinatorica 6(1986),335-354.
[13] G. A. Freiman, On two and three-element subsets of groups, Aeq. Math.
22(1981),140-152.
[14] Z. Fiiredi, Maximum degree and fractional matchings in uniform hypergraphs,
Combinatorica 1(1981),155-162.
[15] Z. Fiiredi, Intersecting designs from linear programming and graphs of diameter
two, Discrete Maths 127(1994), 187-207.
[16] A. Gardiner and C. E. Praeger, Transitive permutation groups with bounded
movement, J. Algebra 168(1994), 798-803.
[17J A. Gardiner and C. E. Praeger, Bounds on orbit sizes for permutation groups
admitting subsets with restricted movement, (preprint, 1995).
[18] B. Ganter, J. Pelikan and L. Tierlinck, Some sprawling systems of equicardinal
sets, Ars Combin. 4(1977), 133-142.
[19] A. S. Hedayatt, G. B. Khosrovshahi and D. Majumdar, A prospect for a general
method of constructing t-designs, Discrete Appd. Math. 42(1993),31-50.
[20J M. Herzog, P. Longobardi and M. Maj, On a combinatorial problem in group
theory, Israel J. Math. 82(1993),329-340.
[21] G. B. Khosrovshahi and C. E. Praeger, On the intersection problem for com-
binatorial designs, (in preparation).
[22] P. Longobardi and M. Maj, The classification of groups with the small squaring
property on 3-sets, Bull. Austral. Math. Soc. 46(1992), 263-270.
[23] A. Mann and C. E. Praeger, Transitive permutation groups of minimal move-
ment, J. Algebra, to appear.
[24] B. H. Neumann, Groups covered by permutable subsets, J. London Math. Soc.
29(1954), 236-248.
[25] P. M. Neumann, The structure of finitary permutation groups, Arch. Math.
(Basel) 27(1976), 3-17.
[26] P. M. Neumann, A combinatorial problem in group theory, (unpublished manu-
script, 1989).
[27] P. M. Neumann and C. E. Praeger, An inequality for tactical configurations,
Bull. London Math. Soc. (to appear).
[28] C. E. Praeger, On permutation groups with bounded movement, J. Algebra
144(1991),436-442.
The Separation Theorem for Group Actions 219

[29] C. E. Praeger, Restricted movement for intransitive group actions, Proceedings

of Groups Korea 1994, (to appear).
[30] C. E. Praeger, Movement and separation of subsets of points under group ac-
tions, (preprint, 1995).
[31] A. P. Street and D. J. Street, Combinatorics of Experimental Design, Clarendon
Press, Oxford, 1987.
 Permutation Groups Whose Subgroups Have Just
Finitely Many Orbits

Dugald Macpherson

School of Mathematical Sciences

Department of Pure Mathematics
University of Leeds
Leeds LS2 9JT
England

1 Introduction
In this note we answer a question of Peter Neumann, based on some earlier more
general questions of R. Zimmer concerning actions of arithmetic groups and Lie
groups on manifolds. The main question of Zimmer is the following.

Which groups can act on an infinite set so that every element of infinite
order generates a cyclic subgroup which has only finitely many orbits
on the set?

We prove the following theorem.

Theorem 1.1 The free group F2 on two generators has a faithful permutation rep-
resentation on N in which every non-trivial subgroup has just finitely many orbits.
Furthermore, for every non-identity wE F 2 , w is a power of some element which
induces a single infinite cycle on a cofinite subset of N.

It is not hard to sharpen the construction of the above group slightly in order
to ensure that the action of F2 is highly transitive; that is, that it is k-transitive for
all positive integers k. The details of this strengthening are omitted. I do not know
if the permutation representation of F2 constructed in the proof of Theorem 1.1 is
already highly transitive. However, by the remark following Proposition 3.6 below,
the group F2 is oligomorphic; that is, it has finitely many orbits on N k for each
positive integer k.
The theorem is proved in Section 2. In Section 3 we discuss Zimmer's question
in more generality. In particular we explain why our theorem answers a natural
extension of Zimmer's question.

221
W. C. Holland (ed.), Ordered Groups and Infinite Permutation Groups, 221-229.
© 1996 Kluwer Academic Publishers.
222 Dugald Macpherson

I am very grateful to Peter Neumann, both for encouraging me to publish this

note, and for allowing me to include the material in Section 3. The latter is entirely
due to him.

2 Proof of the theorem

In the proof below we shall use letters X, Y for the generators of the free group
F2, letters a,b,c,d,e,j,m,n,p,q for elements of the set N on which F2 acts (and
simultaneously as indices of certain words), and U,V,W for words in x,y,x-1,y-l.
We use K,L,M,N,S,Tas exponents of certain words. For aEN and wEF2 , we
denote by a W the image of a under w.

Proof of Theorem 1.1. Let F2 be the free group on generators x, y. Let W be the
set of reduced words (in x,y,x-1,y- 1) in F2. Define a relation ..... on W by putting
u""v if there are Wl,W2EW and M,NEZ\{O} such that W11 U M Wl=W;-lVNw2'
Then"" is an equivalence relation. Let {Wi: i E N} consist of exactly one element,
of least possible length, from each ",,-class other than {I}. By the minimality
assumption, no element Wi can be a proper power in F 2 ; and each word Wi is a
reduced word, not conjugate to any shorter word.
To give the required permutation representation of F2 on N, we specify the
actions of x and Y on N point-by-point. We shall build the two permutations x and
y, and then verify that the subgroup of Sym(N) generated by x and y is indeed
the free group F2 on x, y and that the permutation representation of F2 has the
required properties. The construction of the permutations x and Y is in five kinds
of steps. At any of these steps, each wE F2 will have been defined on a finite subset
of N. In particular, at any of these steps w, regarded as a finite partial permutation
of N, will have a finite number (possibly zero) of finite cycles on N, together with
some 'incomplete' or 'partial' cycles. The key condition which we must preserve in
the construction is the following.

(*) If m;::: 5n, then at step m we do not create any new finite cycles of Wo, ... , W n .
The permutations built immediately before step k in the construction of x and
y will be denoted Xk and Yk respectively. In the definitions below, we sometimes
formally treat x k and Yk as sets of ordered pairs. Pick a bijection F: N ---+ N 3 . Put
xo=Yo=0.
Step 5n. Let k:= 5n. If n E domxk, put Xk+l := Xk. Otherwise, choose mE
N \ (domxk Uranxk UdomYk UranYk) and put Xk+l:= Xk U {(n, m)}.
Step 5n + 1. Let k := 5n + 1. If n E ran x.\;, put Xk+l := Xk. Otherwise, pick
mEN\ (domxk Uranxk UdomYk UranyA;) and put Xk+l :=Xk U{(m, n)}.
Step 5n + 2. Let k:= 5n + 2. If n E domYk, put Yk+l := Yk. Otherwise, pick
mEN\ (domxk UranxkdomYk UranYk) and put Yk+l :=Yk U {(n,m)}.
Permutation Groups Whose Subgroups Have Just Finitely Many Orbits 223

Step 5n+3. Let k:= 5n+3. If n E ranYk, put Yk+I:= Yk. Otherwise, pick
mEN\ (domxk Uranxk UdomYk UranYk) and put Yk+! :=Yk U {(m, n)}.
We must check that each of these four kinds of steps preserves property (*). In
fact, no element Wi acquires a new finite cycle after one of these steps. For example,
consider the case k:= 5n, where Xk+! is a proper extension of Xk with n Xk +, = m.
Suppose that for some lEN and wE{w;:iEN}, IW is defined after Step k, but not
before it. Then, by the choice of m, either 1= m and w begins with X-I, or IW = m
and wends with x. Since w is of least length in its ......-cl ass , at most one of these
can occur. After Step k, in the first case lW-' is not defined, and in the second case
Iw 2 is not defined. Either way, I does not lie in a finite cycle of w, as required.

Step Sn+4. Put k:= 5n+4. Also, let F(n) = (p, q, r). If one of p, q lies in a finite
cycle of w r , or if p, q are in the same partial cycle of w r , do nothing, that is, put
= =
Xk+1 Xk and Yk+! Yk· Otherwise, we join the partial cycle of Wr containing p to
that containing q, as follows.
We first introduce some notation. If wE W, let I( w) be the sum of the absolute
values of the exponents of x and Y in w, that is, the length of w. Let I; := I( wd
for each i E N. If a E Nand wE W with lew) = I, then a w-path at a is a sequence
aI, ... , al such that for each initial segment U of w, if I( u) = j"? 1 then aU = aj. We
shall from now on write w for Wr and I for Ir . When we say that aU is defined for
some a EN and U E W, we mean that it it defined before Step k.
Let SEN be maximal such that pws is defined, and put pWs = a. Let Ul be a
maximal initial segment of w such that aU' is defined, and put aU, =a'. Let TEN
-T -T
be maximal such that qW is defined, and put b = qW . Also, let V2 be a maximal
final segment of w such that bV :;' is defined, and put b':= bV :;'. Let U2, VI be the
= =
unique words in x,y,x-l,y- 1 such that w UIU2 VIV2 (so both U2 and VI are
non-empty, and could possibly equal w). Let mi=l(ui) and n;=I(vi) for i=1,2.
Finally, put
r :=N\ ({p,q}Udomxk Uranxk UdomYk UranYk).
Put N:=(Max{lo, ... ,ln}+2)2. The elements Xk+!,Yk+l will be the least exten-
sions of x k, Yk so that after Step k we have a sequence

from N with the following properties.

( a) The above sequence is a U2wN vI-path at a'.
(b) The Ci, di,j, ei are all chosen to be distinct elements of r apart from en" which
equals b'.
We shall now verify that condition (*) is not violated at this step. So let Wi E
{wo, ... ,wn }, with l(w' ) =1'. We must check that no new finite wi-cycle has been
created, so suppose for a contradiction that there is a new finite wi-cycle. The key
point in the argument below is the following observation.
224 Dugald Macpherson

(**) There is fEN lying in the new finite cycle, and some subword Vi of Wi such
that r' E r; and furthermore, we may choose f and Vi so that r' = Cl, and
so that, if l' EN is large enough, then the sequence

is a path at Cl for an initial subword of wiL' or of (w , - 1 )L'.

This follows from the definition of r and the fact that Wi is of minimal length in
its ~-class. The observation enables us, by the choice of the ci,di,j, and ei, to form
equations involving wand Wi.
Case (i). w'=w.
In this case, by (**), there is i with 1 ~ i ~ I such that the new cycle includes
each dj,i, and either (dj,i)W =dj+l,i for each j (1 ~j ~N -1) or (dj,i)W =dj-l,i for
each j (2 'So.j 'So. N). Suppose first that i < t. The new finite cycle of w includes d2 ,i.
Let u be the initial segment of w of length i (so u # 1, w), and suppose that w = uv.
Then, as noted above, (d 2 ,i)WE{d 1 ,i,d3 ,;}. If(d 2 ,i)w=d3 ,i then the words uw and
wu are equal, and if (d 2 ,i)W = dl,i then the words uw- l and wu are equal. The
latter is clearly impossible; for in this case, if u starts with z E {x, x- l , y, y-l} then
so does w, so w- 1 ends with z-1, so u ends in z-1. Repeating the argument with
the second entry in u, and continuing in this way we find that u is not reduced. In
the former case it follows that w is equal (as an element of F 2 ) to a proper power
of u, contradicting the minimality of w in its ......-class.
It follows that i=l. Now (d j ,i)w=dj+1,i for each j=1, ... ,N-l, and hence the
new cycle includes p and q. However, before Step 5n+4 p and q were in different
w-cycles. Let M be least such that qW M = P after Step k. Then if LEN is largest
such that pw- L is defined before Step k, it follows as in (**) that pw-(L+l) E r (after
Step k). Hence, using (**) and the last paragraph, there is J{ < M such that, after
-K
Step k, p'" = dj,1 for some j E {I, ... , N}. It follows that w does not define a
partial permutation, since dj,1 occurs twice in the cycle containing p.
Case (ii). Wi # W.
By (**), some element of r occurs in the new wi-cycle. In fact, if fEN is on
this cycle and J{ is very large, then all elements of r \ {f} occur in the union of a
W 'K -path at f and a W , - K -path at f. By the pigeonhole principle, as N > (I' + 2)2
there are distinct i, j with 1 ~ i, j ~ Nand s with 1 ~ s ~ Nand MEN such that
1M
(ds,i)W =ds,j.

Let u be a proper initial segment of w of length s. Then u( w')M = wi -i u . It follows

that w ...... Wi, which is a contradiction.
This completes the construction of the Xk and Yk, for each step k. Now put
x:= U(Xk: k EN) and Y :=U(Yk: k EN). By the steps ofform 5n, 5n+ 1, 5n+2, 5n+3,
 Permutation Groups Whose Subgroups Have Just Finitely Many Orbits 225

x and yare permutations of N. We now verify the assertions of the theorem. If

wE Wand w is not the empty word, then w or w- 1 is conjugate to a proper power
of some w r , so we may suppose that w := w r , say. Then the finite cycles of ware
just those defined before Step 5r, so w has finitely many finite cycles. In particular,
w induces a non-identity permutation, so the group generated by x and y is free.
Furthermore, if ~ is the union of the finite cycles of w, and p, q ¢~, then there is
nEN with F(n)=(p,q,r). At Step 5n+4, we ensure that p and q are in the same
cycle of w r • Hence Wr induces a single infinite cycle on N \~, as required.

3 Background to the main theorem

Theorem 1.1 was motivated by the following collection of results, all due to Peter
Neumann in response to the questions of Zimmer.

We adopt the following terminology. Let G be a permutation group on a set O.

We say that (G,O) has property Z if 0 is infinite and every element ofG of infinite
order has only finitely many cycles (including cycles ofJength 1, that is, fixed points)
in O. We say that it has property ZTF if 0 is infinite and every non-identity element
has only finitely many cycles. Evidently, any group with property ZTF is torsion-
free and has property Z; conversely, any torsion-free group with property Z has
property ZTF. Furthermore, a subgroup of a group with either of these properties
has that property. We remark too that if G has property ZTF then it is faithful
on each of its infinite orbits. Observe that the permutation representation of F2 in
Theorem 1.1 has property ZTF.
The first lemma provides examples of groups with property Z.

Lemma 3.1 Suppose that G is infinite and cyclic-by-finite. Then any faithful per-
mutation representation of G with only finitely many orbits has property Z.

Proof. Any element of G of infinite order generates a subgroup of finite index,

and therefore has only finitely many orbits in each G-orbit.

In particular, the regular action of any infinite cyclic-by-finite group has pop-
erty Z. The next lemma gives a partial converse to this.

Lemma 3.2 If (G, 0) has property ZTF and G acts regularly on 0 then G is infinite
cyclic.

Proof. Let 9 be any non-identity element of G. Since the cyclic group (g) has
only finitely many orbits in 0 it has finite index in G. Therefore its core (the
intersection of the G-conjugates of (g}) has finite index, and so G is cyclic-by-finite.
It now suffices to quote the following claim, which for completeness we prove.
Claim. Any torsion-free cyclic-by-finite group is cyclic.
226 Dugald Macpherson

Proof of Claim. Let H be a cyclic normal subgroup of finite index in the torsion-
free group G. If C := CG(H) then the centre Z(C) of C has finite index n, say,
in C. The transfer homomorphism T: C ---+ Z(C) is given by the map 9 f-+ gn
(see [6] Exercise 3.5.8(c) and the discussion preceding it). Since G is torsion-free,
T has trivial kernel, so C embeds in Z(C). Hence C is a torsion-free abelian
cyclic-by-finite group, so is itself cyclic. Replacing H by C we may suppose that
=
CG(H) H. Now conjugation by elements of G induces a faithful action of Gj H
on H by automorphisms. Thus either H = G or G j H is cyclic of order 2. In the
latter case, if 9 E G \ Hand h:= g2 then on the one hand g-l hg = h because h is a
power of g, and on the other hand g-l hg = h -1 because h E H. This would entail
h 2 = 1, hence g4 = 1, and then 9 = 1 since G was supposed to be torsion-free. But
that contradicts the choice of 9 as an element not in H. Thus in fact we must have
H = G, that is, G is cyclic.

The next lemma imposes an important restriction on the groups which can have
property Z.

Lemma 3.3 Suppose that n is infinite and (G, n) has property Z. If 9 E G and 9
has infinite order then its centraliser CG(g) is cyclic-by-finite.

Proof. If 9 has k infinite cycles and if no is the set of points of n permuted by

9 in finite cycles then CSym(O) (g) === ZWrSym(k) x Co, where Co is a subgroup of
Sym(no), and Z is the infinite cyclic group. Since no is finite, so is Co. Let C 1 be
the intersection of CG(g) with the base group of the wreath product, so that the
non-trivial orbits of C 1 are the infinite cycles of g. If n 1 is one of these then C 1
must act faithfully on n 1 . Therefore C 1 is cyclic, and so CG(g) is cyclic-by-finite.

Recall that a permutation group G on a countably infinite set n is said to be

oligomorphic if, for every positive integer k, G has finitely many orbits on nk;
equivalently, if, for every finite r c n, G(r) := {g E G: 9 fr= id} has finitely many
orbits on n.
The next result suggests that a non-cyclic group with property ZTF should be
oligomorphic. However, this has not been proved, and the obstruction involves
infinite Frobenius groups. Recall that a permutation group is a Frobenius group if
it is transitive and every 2-point stabiliser is trivial. We shall call a Frobenius group
implausible if

(a) the stabilisers are infinite cyclic; and

(b) the stabilisers have only finitely many orbits.
We call it highly implausible if in addition
(c) it has property ZTF.
Permutation Groups Whose Subgroups Have Just Finitely Many Orbits 227

We now give a little evidence in the direction that highly implausible groups do
not exist. It is due independently to Karolyi, Kovacs and Palfy [2] and to Mazurov
[4].
Proposition 3.4 ([2, 4]) There is no 2-transitive permutation group whose one-
point stabilisers are infinite cyclic.

The next lemma is permutation group-theoretic folklore.

Lemma 3.5 Let (G, D) be an infinite transitive permutation group, and suppose
that G has finitely many orbits on D2. Let a E D, let <P be the union of the finite
orbits of GO!) and suppose that 1<p1 > 1. Then <P is a block of imprimitivity in the
action of G on D.

Proof. This is left as an exercise. I know of no good reference (it is essentially

Exercise 3(ii) of Section 2.2 of [1], and also was mentioned in a lecture course by
Peter Neumann at Oxford in Autumn 1985).

Proposition 3.6 Suppose that highly implausible Frobenius groups do not exist. If
(G, D) has property ZTF then G is either cyclic or oligomorphic.

Proof Suppose this false, and let (G, D) be a counter-example, so that G has
property ZTF but is neither cyclic nor oligomorphic. Since G has property ZTF it
has only finitely many orbits, and since it is not oligomorphic there is some finite
sequence whose stabiliser has infinitely many, so is the identity. Therefore we can
choose a sequence 0'1, ... , ak, ak+1 which is such that Gal, .,ak has only finitely many
orbits on D, but Gal, ... ,ak,ak+l has infinitely many. We may choose the sequence
so that, in addition, for each i = 1, ... , k, the element ai+1 is in an infinite orbit of
Gal, ,a, (and 0'1 is in an infinite G-orbit). Now Gal, ... ,ak has property ZTF on its
orbit containing ak+1, which is infinite. It acts faithfully and regularly on this orbit
and therefore by Lemma 3.2 it is cyclic.
For O<i<k+1
_ _ define G(i):=G Q'l,···,O'.' so that G=G(O»G(l» ... >G(k»
G(k+ 1 )={1}. Since G is assumed not to be cyclic, ki=0. Furthermore, G(k) has in-
finite index in G(k-1), so G(k-1) is not cyclic. In addition, G(k-1) has property ZTF
and is not oligomorphic. Therefore we may assume, without loss of generality, that
k = 1. Let r be the G-orbit of 0'1. Observe that G(1), being abelian, acts regu-
larly (and faithfully) on each of its infinite orbits on r. Let <P be the union of the
finite orbits of G(1) which lie in r. Since G(l) has only finitely many orbits, <P is
finite. Furthermore, by Lemma 3.5, <P is a block of imprimitivity of G on D. Let
p be the corresponding congruence (G-invariant equivalence relation) on r. Any
element of G which acts trivially on the quotient rip fixes all the p-classes, and
since these are finite, has infinitely many cycles, hence is 1. Thus G acts faithfully
on rip. Pick /31 Er\<p (so G al ,{31 = 1). Let a:=p(ad, let /3:=p(/3d thought of
as elements of Dip, and let H:= G «,{3' In the action on r the sets p( ad and p(/3d
228 Dugald Macpherson

are H-invariant, and so Ha"P, has finite index in H. But Ha"P,::; GCXl,Pl = {I}.
Therefore H is finite, and so, since G is torsion-free, H = {I}. This means that G
acts as a Frobenius group with cyclic stabiliser on r j p. Since, as is easy to see,
(G,0.jp) also has property ZTF, it is a highly implausible Frobenius group. This
contradicts our assumption that these do not exist.

Remark. Let X be a subgroup closed class of groups, and suppose that no

member of X has a faithful permutation representation as a highly implausible
Frobenius group. If GEX and (G,0.) has property ZTF then G is either cyclic or
oligomorphic. For, in the proof of Proposition 3.6 we only needed that subgroups of
G had the same properties as G. In particular, if G is a non-abelian free group acting
faithfully as a permutation group with property ZTF then Gis oligomorphic (so the
action in Theorem 1.1 is oligomorphic). Theorem 1.1 shows that the oligomorphic
case in Proposition 3.6 really does occur.

Proposition 3.7 Suppose that (G,0.) is transitive and has property ZTF. Suppose
furthermore that for every natural number k there exists a non-identity element of
G that has at least k fixed points and just one infinite cycle. Then G is highly
transitive on 0..

Proof. First, observe that if (H, I:) is a transitive permutation group, and H
contains an element h with at least one fixed point and with an infinite cycle which
contains all but finitely many points of I:, then H is primitive on 0.. This is a
slight variation on a result from [3]. To prove it, let 0: be a fixed point of h, let
r be the set of points on the infinite cycle of h, and let p be an H -congruence on
I:. If p( 0:) n r = 0 then, as I: \ r is finite, p( 0:) is finite, so all the p-classes are
finite. In this case the restriction of p to r must be trivial, and, since only finitely
many p-classes are disjoint from r, it follows that p is trivial. If, on the other hand,
p(0:)nr:;i:0, then, applying powers of h, we see that rCp(o:). In this case peo:)
has finite complement in I: and is infinite, so is the whole of I:, that is, p is the
universal relation. Hence H is primitive.
Let (G,0.) be as in the statement of the proposition, and let (tE0.. By transi-
tivity and our other assumption, there is a non-identity element 9 of G a that has
at least one fixed point, finitely many finite cycles, and one infinite cyde. By what
we have just proved, G is primitive. Furthermore, G a has one infinite orbit and
finitely many finite orbits. By Lemma 3.5, as (G,0.) is primitive, G n has no finite
orbits on 0.\ {a}, so G a is transitive on fJ\ {a}. In particular, Gis 2-transitive.
Now G a certainly satisfies the same hypotheses as G. We can therefore assume as
inductive hypothesis that G n is k-transitive, and get that G is (k+ I)-transitive.
Thus by induction G is k-transitive for every finite k, that is, it is highly transitive.

We record three questions on property ZTF.

Question 3.8 Is there a (highly) implausible Frobenius group?

Permutation Groups Whose Subgroups Have Just Finitely Many Orbits 229

Question 3.9 Is every primitive group which has the property ZTF and is also
oligomorphic necessarily highly transitive?

Question 3.10 What classes of groups other than free groups have oligomorphic
actions with property ZTF?

Finally, I remark that there is now quite a wide literature on possible cycle types
of a particular permutation (rather than all permutations) in infinite permutation
groups. For example, there are several naturally occurring permutation groups in
which the realisable cycle types are classified, and in particular some of these have
permutations acting as a single infinite cycle (see [5,7,8,9] for more on this). Also,
in [3] there are results restricting the possible cycle types in a primitive permutation
group which is not highly transitive.

References
[1] P . .J. Cameron, Oligomorphic Permutation Groups, London Math. Soc. Lec-
ture Notes Series 152, Cambridge University Press, Cambridge, 1990.

[2] Gy. Karolyi, S . .J. Kovacs, P. P. Palfy, Doubly transitive permutation groups
with abelian stabiliser, Aequationes Math. 39(1990), 161-166.

[3] H. D. Macpherson and C. E. Praeger, Cycle types in infinite permutation

groups, .J. Algebra, to appear.

[4] V. D. Mazurov, Doubly transitive permutation groups, Sibirsk. Mat. Zh.

31(1990), 102-104 (in Russian).

[5] P. M. Neumann, Automorphisms of the rational world, .J. London Math. Soc.
32(1985),439-448.

[6] W.R. Scott, Group Theory, Prentice-Hall, Englewood Cliffs, 1964.

[7] J. K. Truss, The group of the countable universal graph, Math. Proc. Cam-
bridge Philos. Soc. 98(1985), 213-245.

[8] J. K. Truss, The group of almost automorphisms of the countable universal

graph, Math. Proc. Cambridge Philos. Soc. 105(1989),223-236.

[9] J. K. Truss, Generic automorphisms of homogeneous structures, Proc. London

Math. Soc., to appear.
 Automorphisms of Quotients of Symmetric Groups

J. L. Alperin Jacinta Covington

Department of Mathematics Department of Mathematics

University of Chicago Royal Melbourne Institute of Technology
5734 University Avenue GPO Box 2476V
Chicago Illinois 60637 Melbourne, Victoria 3001
U.S.A. * Australia t

Dugald Macpherson

Department of Pure Mathematics

University of Leeds
Leeds LS2 9JT
England

Abstract
We describe the automorphism group of the quotient of the symmetric
group on a count ably infinite set by the group of finitary permutations. We
show that the outer automorphism group is infinite cyclic. In the proof, we
examine the structure of centralizers in this quotient.

1 Introduction
In this paper, S will denote Sym(O), the full symmetric group on a countably infinite
set O. For any g E S, define the support of g to be supp(g) = {a E 0: ag i=- a}
and the fixed point set to be fixg = {a E 0: ag = a}. It is well-known (see
Scott [6] or originally Baer [1]) that S has precisely two proper non-trivial normal
subgroups. These are FS(O) := {g E S: I supp(g) I < ~o} and Alt(O), the subgroup
of index two in FS(O) consisting of even permutations. Throughout this paper, we
let G := S/FS(O). This group is simple. The main goal of this paper is to describe
Aut(G).
·This research was partially supported by a grant from the National Science Foundation.
tThis research was partially supported by ARC grant A68830987.

231
W. C. Holland (ed.), Ordered Groups and Infinite Permutation Groups, 231-247.
© 1996 Kluwer Academic Publishers.
232 Alperin, Covington, and Macpherson

Theorem 1.1 Let G = Sym(n)/FS(n), where n is a countably infinite set. Then

the ouier automorphism group Out( G) is infinite cyclic.

To see why G admits an infinite cyclic group of outer automorphisms, observe that
if n is identified with N, then the shift n ....... n + 1, acting on G 'by conjugation up
to finitely many mistakes', induces an outer automorphism of infinite order.
We give in Section 2 a more formal description of the infinite cyclic group of
outer automorphisms. In Section 3 we describe in Proposition 3.2 and Theorem 3.3
the structure of centralizers in G, and thereby show that certain conjugacy classes
of G are invariant under Aut(G). We believe that these results are of independent
interest, as the centralizers do not have as simple a structure as might first be
imagined. In Section 4, we show that there are no outer automorphisms other than
those described in Section 2, and thereby prove Theorem 1.1.
We frequently refer to elements of S by their cycle type. The set of cycle lengths
is the set of cardinals {n: 1 ::; n ::; w}, and a nontrivial cycle has length n > 1. If
T is a set of cycle lengths, we say that 9 has cycle type I1(n an : nET) if g has
exactly an cycles of length n for each nET, (0 ::; an ::; w), and no others. Observe
that if g, h E S have cycle types 1w. I1 (nan: 1 < n ::; w) and 1w. I1 (n bn : 1 < n ::; w)
respectively, then FS(n)g, FS(n)h are conjugate in G if and only if their cycle types
differ finitely: that is, if
(a) an = bn for all but finitely many finite cycle lengths n;

(b) a w = bw ;
( c) an = w if and only if bn = w, for 1 ::; n < w.
It is important that here g and h have infinitely many fixed points. As a small
warning, we remark that G has three conjugacy classes of involutions, corresponding
to elements in S of the cycle types 1w . 2w and 2W and 11 . 2W •
If 9 E S then g will be the corresponding element FS(n)g of G, and similarly,
given g E G, we use 9 to denote any preimage of it under the natural surjection
from S to G. We shall use t:.. to denote symmetric difference, and write U == V if
U t:.. V is finite. Also, we extend the notion of support to elements g E G. If 9 E S
has support r ~ n, then g has support

supp(g) = {~ ~ n: Irt:..~1 < w}.

Thus supp(g) lies in the Boolean Algebra B = pow(n)/I, where I is the ideal of
finite subsets of n. Observe the distinction between supp(g), which is a subset of
n, and supp(g), which is an element of B. All wreath products whose top group is
a permutation group of infinite degree will be unrestricted, and similarly all direct
products will be unrestricted (that is, will be the full Cartesian product). Finally,
recall that a moiety r of a set n is a subset satisfying Irl = In \ rl; for a countably
infinite set n this is an infinite coinfinite subset.
 Automorphisms of Quotients of Symmetric Groups 233

The paper by J .K. Truss in the present volume sketches an alternative proof of
Theorem 1.1, due to Truss and Shelah, in which the Boolean algebra B is shown to be
first-order interpretable in G. Their approach seems to be more powerful, since they
obtain some generalisations of Theorem 1.1 for symmetric groups of uncountable
cardinality. The strongest results concerning the interpretability of B in G were
obtained by M. Rubin in Theorem 4.2 of [5]. There, he showed the following. Let
ft < K ::; >. + be infinite cardinals and G >'K./1 denote the group of permutations (of
support of size less than K) of A, modulo the normal subgroup of permutations of
support of size less than ft, and B>'I</1 be the corresponding Boolean algebra. Then
B>'''/1, its boolean algebra structure, and the action of G >'1</1 on it, is first-order
interpretable in the abstract group G >'1</1' and furthermore this interpretation is
uniform, that is, the formulas in the interpretation are independent of the choice
of >., K, ft. It seems, though, that the step from such results to a description of
Aut( OG >'''/1 is never immediate. Our approach yields en route the group-theoretic
information about centralizers.

Acknowledgement. We would like to thank both Peter Neumann and John Truss
for a number of very helpful remarks.

2 The group of near symmetries

In this section we describe an infinite cyclic group of outer automorphisms of G =
Sym(0)/FS(0), where 0 is a count ably infinite set. In the rest of the paper, we
will show that this group is the whole outer automorphism group.
Define the set of near bijections of 0 to be

NB(O) = {f: f is a bijection 0 1 ~ O 2 , where 0 1 , O2 are cofinite subsets of 0}.

Clearly, S ~ NB(O). Extending our notation for sets, we define an equivalence

relation == on NB(0) by the rule

h == h if there is a cofinite 0' ~ 0 such that h Irv= h Irv .

It is easily seen that == is indeed an equivalence relation. We define NS(0), the set
of near symmetries of 0, to be the set of equivalence classes of this relation. This
set naturally forms a group induced from composition of maps.
The natural projection from S to NS(0) is a group homomorphism with kernel
FS(0). Hence we have an exact sequence

FS(0) --+ S --+ NS(0).

Thus G ::; NS(0). The elements of G are the equivalence classes of permutations.
For f E NB(O) with 0 1 , O2 as above, we define the index of f to be
234 Alperin, Covington, and Macpherson

Lemma 2.1 The map ind: NB(O) --+ Z induces a mapping ind: NS(O) --+ Z which
is a homomorphism onto (Z, +) with kernel G.

Proof. First note that if 0' is a cofinite subset of 0 then for any f E NB(O) we
have ind f = ind (I 10 /). Take f, 9 E NB(O) with f == g. We want to show that
ind f = =
ind g. There is a cofinite subset 0' of 0 such that f 10/ 9 10/. But then
clearly
indf = ind(l 10 = ind(g 10/) = indg.
1)

Thus ind is a well-defined function on NS(O).

We now show that ind (lg) = indf + indg. Restricting the domains of f and
9 if necessary, we may assume that dom 9 is equal to ran f, the range of f, that is,
we have f: 0 1 --+ O2 and g: O2 --+ 0 3 . Then
ind (lg) 10 - 0 3 1-10 - 011
10 - 03 1- 10 - 0 2 1+ 10 - 0 2 1- 10 - 011
ind 9 + ind f.
Now the kernel of this homomorphism consists of all bijections f: 0 1 --+ O2 such
that 10 - 011 = 10 - 0 2 1. Any such map is ::=-equivalent to a permutation. Thus
ker(ind) = G. 0

From this lemma it follows that G is a normal subgroup of NS(O) and we have
an exact sequence
1 - + FS(O) -+ S -+ NS(O) -+ Z -+ 1.
Lemma 2.2 CNS(O)(G) is trivial.

Proof. Let h be a near-bijection representing an element of CNS(O)(G). Then

for every permutation 9 E S, we have h- 1 gh ::= g. Now, for any 9 E S, the sets
supp(h-lgh) and (supp(g»h have finite symmetric difference, where supp(h- 1 gh)
is the the set of CY E 0 such that cyh- 1 gh is defined and not equal to CY. It follows
that supp(g) ::= supp(g)h for any 9 E S. Since any set <J> C 0 is the support of some
permutation, we have <J>h ::= <J> for all <J> ~ O. But this is possible only if h has finite
support, that is, h::= id. 0

Proposition 2.3 NS(O) induces an infinite cyclic group of outer automorphisms

ofG.

Proof. Since G::::! NS(O), we see that NS(O) induces a group of automorphisms
of G by conjugation. Since CNS(O)(G) is trivial, the outer automorphism group
induced by NS(O) is NS(O)/G, which is isomorphic to Z. 0

Remark. Let 0 = {CYj: i < w} be any enumeration of O. A typical generator

of NS(O)/G is the equivalence class of the near-bijection u: 0 --+ 0 defined by
CYj f--+ CYj+l (for i < w).
Automorphisms of Quotients of Symmetric Groups 235

3 Structure of centralizers in G
In this section we give partial information on the structure of centralizers in G.
From this, it will follow that Aut(G) induces an automorphism of 8, and hence
that Aut(G) fixes certain conjugacy classes of G.
For the first four lemmas we denote by g a fixed element of G, and choose a
fixed 9 E S such that g = gFS(Q). We suppose that 9 has cycle type IT(in,: i E T),
where T ~ N U {w}. We define an equivalence relation F on Q whose equivalence
classes are listed below:

• The union of all the infinite cycles of 9 is an F-class Fw.

• If 1 :S i < wand 9 has infinitely many i-cycles, then the union of the i-cycles
of 9 is an F-class F i .

• The remaining points of Q form a single F-class Fa, called the dregs.

We define the strong centralizer of g in G, to be

SCe(g) := {Ii E Ce(g): h fixes each F-class for some representative h of Ii}.

Note that, even though F depends on g, this definition does not depend on the
choice of g. For, let g' be another preimage of g, and let F' be the corresponding
equivalence relation. Take h E S such that Ii E SCe(g). Now, P and F differ only
on a finite set ~ of points. There is a finitary permutation k fixing each F-class
such that hk E S(Ll.)' Then hk fixes each P-class.

LeIllIlla 3.1 Let g, g be as above. Then the group Ce(g)/SCe(g) is free abelian
(where the trivial group is regarded as free abelian).

Proof Let {Fi: i E I} be the set of F-classes, and let K := EB(ZJ;: i E I) be free
abelian with {k i E I} as basis. We define a homomorphism x: Ce(g) ---+ K as
follows: if k E Ce(g) then for each i E I, we define

Xi(k) = IFi k \ Fd - IF; \ Fikl·

Then Xi(k) is well-defined, that is, it does not depend on the choice of the preimages
9 and k. Also put
X(k) := L xi(k)!;.
iEI

Now 9 and k- 1 gk agree on all but finitely many points, so k fixes all but finitely
many F-classes and Xi(k) = 0 for all but finitely many i E I, so X(k) does indeed
lie in the direct sum K. It can be checked that X is a homomorphism, its image is
a subgroup of a free abelian group so is free abelian, and its kernel is SCe(g). 0
236 Alperin, Covington, and Macpherson

Problem 3.1 Find an expression for the rank of CG(g)/SCG(g), in terms of the
cycle type of g.

We will need a considerable amount of information about all centralizers in G.

To find CG(g), we must describe all elements h E S such that [g, h] E FS(O). We
list five types of such elements, and then show that the images in G of elements of
such types generate CG(g).

1. Obvious Centralizer Type These are elements of Cs(g). Conjugation by

these elements fixes g. Note that Cs(g) e:' I1iET(C; Wr Sn.), where C j de-
notes the cyclic group of order i.

2. Finite/Finite Type These elements convert one finite set of finite cycles of g
into another. Let A, B be the unions of supports of finitely many finite cycles
of g, where

• IAI = IBI > 0;

• No cycle in B has the same length as one in A;
• A or B contains the support of a g-cycle of length i only if nj = w, that
is, g has infinitely many cycles of length i.

Let x: A - t B be a bijection and define a permutation h as follows:

• h agrees with X on A;
• If A contains an i-cycle, then h maps F; \ A to F;, respecting the cycle
structure;
• If B contains an i-cycle, then h maps F j to F; \ B, respecting the cycle
structure;
• h fixes all elements of all other F-classes.

Then gh agrees with g except on the finite set B, so [g, h] E FS(O).

3. Finite/Infinite Transfers Conjugation by these elements moves a finite cycle

of g into an infinite cycle. Let ( ... a-2a-laOala2 ...) be an infinite cycle of g.
Let (b 1 b2 ... bk ) be a finite cycle of g from the infinite F-class Fk. Define a
permutation h as follows:

• h fixes aj for i :::; 0;

• a·h
Z = b· for 1 < i < k·
2 __ ,

• aj h = ai _ k for i > k;
• h maps Fk to Fk \ {b 1 , b2 , ... , bk }, respecting the cycle structure;
• h fixes all elements of all other cycles.
A utomorphisms of Quotients of Symmetric Groups 237

Then gh agrees with 9 except on {aD, bd, and so [g, h) E FS(fl).

We refer to the inverses of such elements as infinite/finite transfers.
4. Infinite/Infinite Transfer These are elements which move a single point from
one infinite cycle of 9 into another. Let

be two infinite cycles of g, and let h be the permutation of fl whose only

non-trivial cycle is
( ... L 2L I boa oala2 ... ).
Then the conjugate gh differs from 9 only in that ao lies in the b-cycle of gh .
Hence gh agrees with 9 except on the set {a_I, aD, bo}, so [g, h) E FS(fl).

5. Infinite/Infinite RecOInbination These are elements which cut two infinite

cycles of 9 in half, then reattach the left side of one to the right side of the
other. Let 9 have two infinite cycles as above, and let

h := (aIbl)(a2b2)'"

Then gh agrees with 9 except on {aD, bo }, so [g, h) E FS(fl).

Remark. Elements of types 1,4 and 5 lie in SCG(g), but elements of types 2 and
3 do not.

Proposition 3.2 (a) The group SCG(g) is generated by elements of G repre-

sented by elements of S of types 1,4 and 5.
(b) The group CG(g) is generated by elements of G represented by the above types.

Proof. (a) Let k E S be such that [k,g) E FS(fl), and assume that k fixes setwise
each F-class. We say that two infinite g-cycles

and
( ... L2LIbobIb2 ... )
are initially equal if there are integers M and m such that ai = b; -m for all i < M.
Finally equal is defined similarly.
For all but finitely many of the infinite g-cycles, its conjugate under k is also a
g-cycle. Let ( ... a-2a-IaOaIa2 ... ) be any infinite g-cycle. Then

( ... a_2k,a_Ik,aok,alk,a2k, .. )

is initially equal to some g-cycle, and finally equal to some, perhaps different, g-
cycle. Then we can use infinite/infinite recombinations to reduce to the case where
238 Alperin, Covington, and Macpherson

(ai k) is initially and finally equal to a particular infinite g-cycle. We then use a
series of infinite/infinite transfers, together with finitary permutations to reduce to
the case where k preserves the structure of all the infinite cycles. Using further
finitary permutations, we may suppose that k preserves all finite g-cycles, and it
follows that k E CG(g).
(b) Let k E 5 with [k, g] E FS(O). Observe that Fk6.F is finite for every F-
class F, and that all but finitely many of the F-classes are fixed setwise by k. By
adjusting k by a finitary permutation, we can reduce to the case where, for each
F-class F, either F <;;;; Fk or Fk <;;;; F. Note that since the dregs class Fo contains
only finite cycles and has just finitely many cycles of any length, we must have
Fok = Fo·
Now consider the classes Fi where 1 S i < w. We wish to adjust k to ensure
that F;k = Fi for each such i, for then also Fwk = Fw. So fix i with 1 S i < w, and
suppose that F; C Fik. (The case when F; =:> Fik is similar.) If g has an infinite
cycle then we can use a series of infinite/finite transfers to move cycles of Fik \Fi into
Fw, then a combination of finitary permutations on Fw and finite/infinite transfers
to move appropriate elements of Fw back into Fi . If g has no infinite cycle, we
multiply k by a single element of finite/finite type to reduce to the case where
k fixes all F-classes setwise. We can then use a finitary permutation to ensure
that k respects the cycle structure of finite cycles. After these reductions we have
k E SCG(g), so can apply part (a).
o

We now give a partial structure theorem for centralizers in C. We first define

some normal subgroups of C G (g). Let g E 5 be as above. Let

OC(g) := Cs(g) . FS(O)/FS(O) = {Ii: hE 5, [g, h] = I}

the group of images of elements of obvious centralizer type. Also, define

N1(Tn:= {Ii E OC(g): h fixes Betwise the support of each infinite g-cycle},
N 2 (g) := {Ii E SCG(g) : IDh6.DI < ~o, if D is the support of an infinite g-cycle},
and

These definitions are independent of t.he choice of the representative g.

Theorell1 3.3 (aJ The groups

are each normal subgroups of CG(g), wzth

Automorphisms of Quotients of Symmetric Groups 239

(b) N2(g) n OC(g) = Nl (g). Hence we have the isomorphisms N 3 (g)/OC(g) =='
N2(g)/N 1(g) and N 3 (g)/N2(g) ==' OC(g)/N1(g).

(c) (i) The groups CG(g)/SCG(g) and N 2(g)/N 1(9) are Jree abelian, and the
latter has rank n iJ g has n + 1 infinite cycles (n finite) and ~o iJ g has
infinitely many infinite cycles.
(ii) The quotient OC(g)/ Nl (g) is isomorphic to the symmetric group on the
set oj infinite cycles oj g.
(iii) The quotient SCG(g)/ N 3 (g) is isomorphic to the finitary symmetric group
on the set oj infinite g-cycles.

(d) IJ 9 has more than one infinite cycle then the five normal subgroups listed in
(a) are distinct, and iJ 9 has at most one infinite cycle then they are all equal.

Proof. All parts of this are easy, so we omit details. To show that N 2 (g)/Nl(g)
is free abelian, we argue much as in the proof of Lemma 3.1. This quotient is
generated by infinite/infinite transfers, modulo Nl (g). By Proposition 3.2( a), the
finitary symmetric group SC G(g)/N3 (g) is generated by infinite/infinite recombina-
tions (modulo N3(g)). D

We now define, for any h E S, the group

G(h) := {k : supp(k) ~ fix(h)}.

Observe that this definition is independent of the choice of the element h E S such
that h = h . FS(f2). Also, clearly G(g) ~ CG(g).

Lemma 3.4 (a) Suppose that g E S has infinitely many fixed points. Then G(g)
is isomorphic to G, and is the only normal subgroup oj CG(g) which is iso-
morphic to G.

(b) Suppose that g E 5 has just finitely many fixed points. Then there is no
normal subgroup oJCG(g) isomorphic to G.

(c) Let g, hE 5 and <p E Aut(G), and suppose g<p = h. Then 9 has infinitely fixed
points iJ and only iJ h has infinitely many fixed points, and iJ they do, then
G(g)<p = G(h).

Proof. ( a) Let f := fix(g). Clearly

G(g) = Sym(f) . FS(S1)/FS(f2) ==' Sym(f)/FS(f) ==' G.

Now let N be any normal subgroup of CG(g) isomorphic to G. We shall show that
N = G(g). First observe that G(g) is simple, so for any normal subgroup M of
240 Alperin, Covington, and Macpherson

Ca(g), either G(g)::; M or G(g)nM = 1. In particular, by Lemma3.1, as iI is non-

abelian it must lie in SCa(g). Also, by Theorem 3.3, as G is not embeddable in a
free abelian group or a finite or finitary symmetric group, we must have iI ::; OC(g).
Define N := {h E S: h E iI}. Observe that

Cs(g) = II(Ci WrSni : i E T),

and that 1 E T and n1 is infinite, and that OC(g) = Cs(g) . FS(O)/FS(O). Put

f{:= C 1 WrSn, x II(C~i : i E T,i > 1),

and
L := {k E S : supp(k) ~ fix(g)} ~ Snl = S.
Then L::; K. Put [{ := f{ . FS(O)/FS(O) and L := L . FS(O)/FS(O). Then Land
R are normal subgroups of OC(g) and R / L is abelian. The proof will be complete
if we can show that iI ::; L (since L is simple), so we shall suppose that this is false.
It follows that iI n R = 1; for otherwise iI ::; [{ which is impossible as [{ / L is
abelian and iI is non-abelian and simple.
Now pick hE N\L.FS(O). Then h ¢ f{ ·FS(O), so, up to a finitary permutation,
h induces a permutation of infinite support on the set of non-trivial cycles of g. It is
now easy to choose k E I< so that [h, k] ¢ L· FS(O). Now [h, k) E N n (I< . FS(O) \
L . FS(O», which contradicts the last paragraph.
Part (b) follows from the proof of uniqueness in (a), and (c) follows from (a)
and (b).
o

We will adopt the notation G(g), G(h) later in the paper. Also, from now on t.p
will denote an element of Aut(G).

Lemma 3.5 Let t.p E Aut(G), let g, h, kEG, and assume that g, h, k each have
infinitely many fixed points. Then
(a) supp(g) = supp(h) if and only ifsupp(gt.p) = supp(ht.p),
(b) supp(g) < supp(h) if and only ifsupp(gt.p) < supp(ht.p),
(c) supp(k) = supp(g) I\supp(h) if and only ifsupp(kt.p) = supp(g<p) I\supp(h<p).
Proof. To see (a), observe that supp(g) ::; supp(h) if and only if G(h) ::; G(g),
and by Lemma 3.4( c), this holds if and only if G(ht.p) ::; G(gt.p). But, again by
= =
Lemma 3.4(c), we have G(gt.p) G(g)t.p and G(ht.p) G(h)t.p.
Part (b) follows directly from (a). Part (c) then follows since

supp(k) = supp(g) 1\ supp(h)

Automorphisms of Quotients of Symmetric Groups 241

if and only ifsupp(k)::; supp(g)l\supp(h) and supp(i)::; supp(k) for all i E G such
that supp(i) ::; supp(g) 1\ supp(h). 0

It follows from this lemma that Aut(G) embeds in Aut(B) (via the action of
each automorphism on the set of supports of elements with infinitely many fixed
points). Observe that if supp(g) 1\ supp(h) = 0 then there are representatives 9 of
g and h of h with disjoint supports.
Next, we define the conjugacy classes in which we will be especially interested.
For any subset T ~ N \ {OJ, let

G(T)inf := {g : 9 E S, 9 has cycle type l w . II(n W : nET)}.

Also, define

G(T):= {g: 9 E S,g has cycle type lw. II(nI:n E Tn.

Both G(T)inf and G(T) are conjugacy classes of G. Our goal is to show that they
are preserved by any automorphism of G.
For T = {2} we shall write C(2)inf for G( {2} )inf.

Lemma 3.6 Let <p E Aut(G), and let T ~ N \ {OJ, and g E G(T)inf. Then
g<p E C(T)inf.

Proof First, observe that for any i with 1 < i < wand hE S with infinitely many
fixed points, h has infinitely many i-cycles if and only if

SUpp(hi) < I\(supp(hi) : 1 ::; j < i,jli).

By Lemma 3.5, it follows that if k E S with k = g<p, then for each i with 1 < i < w,
the element k has infinitely many i-cycles if and only if i E T.
Next, again assuming h E S has infinitely many fixed points, note that h has
a single infinite cycle and just finitely many non-trivial finite cycles if and only if
supp(h) '10 and we cannot write h = hIh2 with the supports of hI and h2 disjoint
and not equal to 0 or 1 E B. Hence, given such an element h, there is k E S with a
single infinite cycle and no other non-trivial cycles such that k = h<p. Also, if h E S
is not of this form, then h has at least one infinite cycle if and only if there are
hI, h2 E S such that h = hI h2, the hi have disjoint supports not equal to 0 or 1 in
B, and hI has a single infinite cycle and no other non-trivial cycles. It follows from
Lemma 3.5 that if k E S with k = g<p, then k has no infinite cycles. In fact, this
argument shows that if kt, k2 E Sand ki <p = k2 and ki has infinitely many fixed
points, then ki and k2 have the same number of infinite cycles.
Finally, suppose that h E S has infinitely many fixed points and no infinite
cycles. Then the dregs class Fo of h is finite if and only if, whenever there are
242 Alperin, Covington, and Macpherson

hI, h2 E S such that h = hI h2 with h t , h2 having disjoint non-trivial supports and

hI having finitely many i-cycles for all i (1 < i < w), the elements h2 and hare
conjugate in O. Again, this condition is <p-invariant, so the lemma follows from this
and the previous two paragraphs. 0

Lemma 3.7 Let <p E Aut( 0), and let T be an infinite set of nontrivial finite cycle
lengths. Then G(T) is <p-invariant.

Proof. We will show that G(T) is the only conjugacy class of 0 \ {I} whose
elements g satisfy all the following properties.

(a) there is k E C(T)inf with supp k A supp g=0 such that kg E G(T)inf;

(b) for any moiety R of T, there is no kE G(R)inf with supp k A supp g = 0 and
kg E G(R)inf;
(c) if g = hk where h, k E 0\ {I}, supp hAsupp k = 0 and supp hVsupp k = supp g,
then one of (a) or (b) fails when 9 is replaced by Ii.

By Lemma 3.5, the conditions on supports arc preserved by <p, and by Lemma 3.6,
G(T)inf is <p-invariant, so the conditions (a)-(c) are preserved by <p. Hence, given
the above, G(T) is <p-invariant, as required.
Clearly if 9 E G(T) then it satisfies (a)-( c). Conversely, assume that 9 E
0\ {I} satisfies (a)-(c), and choose a representative g. Property (a) says that 9
has infinitely many fixed points, that all but finitely many of the cycles of 9 have
lengths lying in T, and that 9 has no infinite cycles. Property (b) says that all
but finitely many elements of T arise as lengths of cycles of g. Thus (a) and (b)
imply that the set of cycle lengths of 9 is almost exactly T. If the cycle type of 9
differs more than finitely from 1W • II (n 1: nET), then (c) fails when we let h be the
proouct of exactly one n-cycle of 9 for each nET, and let k be the product of the
remaining g-cycles. 0

4 Proof of the Theorem

In this section, we prove Theorem 1.1, that is, that every outer automorphism of
o is one of the outer automorphisms induced by a near symmetry as described in
Section 2.
Let <p be any automorphism of O. We must find f'P E NB(!.l) such that the
action of <p on G is that of conjugation by f'P' In the introduction, we defined the
support of an element of 0, and in Lemma 3.5 we showed that. <p respects supports
of elements of 0 and the 'containment' relation induced on supports by the subset
relation on!.l. We also showed that certain conjugacy classes of 0 in which we
are interested are respected by <p, that is, are not fused by <p. We now pick a
particular conjugacy class G(T) of 0, a subset D of unordered pairs from G(T),
Automorphisms of Quotients of Symmetric Groups 243

and an equivalence relation £ on 1), and show that C(T), 1) and £ are invariant
under <p. Elements of 1) / £ will be naturally identified with ordered moieties of fl,
and this enables us to define f<p E NB(fl) corresponding to <p. We show that f<p
and <p agree in their action on one non-trivial conjugacy class of G. Then since G
is a simple group, this conjugacy class generates G and so the actions of f<p and <p
agree on the whole of G, as required.
Fix a set T = {ni: i < w} of finite nontrivial cycle lengths such that for all i < w
we have ni > 2 I":j <i nj. We note without proof the following easy lemma.

Lemma 4.1 Let T be as above. Then

(aJ if a, b, c E T then c # 2a + b - 2,-

(bJ if a, b, c E T then 2a # 2b + c - 1,-

(cJ ifa,b E T then a # 2b-I,-

(dJ if a, b, c, d E T with a + b = c + d then {a, b} = {c, d};
(eJ if a, b, c, d E T with 2a + b = 2c + d then a = c and b = d.

We say that {g, h} is a matched C(T)-pair if:

(a) g, h have cycle types IW . nw: i E T);

(b) supp(g) U supp(h) is a moiety of fl;

(c) for each nET, the n-cycle of 9 meets supp( h) in a singleton lying in the
n-cycle of h.

Next, let 1) be the set of all unordered pairs {g, Ii} for which there are corre-
sponding elements g, hE S which form a matched C(T)-pair. We say that {g, h} is
a matched cover of {g, Ii}.
If {g, Ii} E 1), we will write supp(g, h) for supp(g) U supp(h). If {g, h} is a
matched cover of {g, Ii}, let the spine o-(g, h) of {g, h} be the w-sequence from fl
whose ith entry is the point in which the ni-cycles of 9 and h intersect. Then the
ribcage of {g, h} is the set supp(g, h) \ (supp(g) n supp(h)), that is the elements of
the supports of 9 and h which are not in the spine.

Lemma 4.2 The set 1) defined above is <p-invariant for any <p E Aut(G).

Proof. We show that if g, h E G then {g, h} E 1) if and only if the following

conditions all hold:

(a) g, hE C(T);
(b) supp(g) V supp(h) # 1;
244 Alperin, Covington, and Macpherson

(c) for all k E C(2)inf such that supp(k) = supp(y) A. supp(h) we have yk, hk E
C(Q) for some subset Q of {ni + nj: ni, nj E T};
(d) yh E C(R), where R = {2n - 1: nET}.
Here, part (b) says that supp(g) U supp(h) is a moiety of O. Part (c) ensures that
for some g, hE S representing y, h, every non-trivial cycle of 9 intersects supp(h) in
at most a singleton, and similarly with g, h interchanged. For, if the ni-cycle of 9
intersects supp(k) in a singleton, which k pairs with a singleton from the nj-cycle of
g, then gk contains an (nj +nj )-cycle. On the other hand, if, say, there are infinitely
many cycles of 9 which each intersect supp(h) in more than one point, then k can
be chosen so that gk contains an infinite cycle. We leave the details to the reader.
If the m-cycle of 9 meets the k-cycle of h in a singleton, then they form an
(m+k-1)-cycle in the product gh. By Lemma4.1(d), ifm, k, nET then m+k-1 =
2n -1 if and only if m = k = n. Thus part (d) ensures that these cycles are matched
up correctly, that is, the n-cycle of 9 meets the n-cycle of h.
By Lemmas 3.5, 3.6 and 3.7, the conditions in (a) to (d) are preserved by any
automorphism. So 1) is if-invariant. 0

Next, we define an equivalence relation £. on 1). If {Yl, hd, Vh h2 } E 1), then

we say that they are equivalent under £. if they have matched covers {gl, hI} and
{g2, h 2} respectively, such that l7(gl' hI) = l7(g2' h2)'
Lemma 4.3 Let if E Aut(G). Then the equivalence relation £. is if-invariant.

Proof. We say that gl, g2, g3, g4 E S form a matched quadruple if

(a) {Yi, Yj} E 1) and has {gj, gj} as a matched cover for all 1 :::; i <j :::; 4;

(b) all sequences l7(gj,gj), (1 :::; i < j:::; 4) are equal;

(c) U{SUpp(gi): I:::; i:::; 4} is a moiety of 0;
Claim: The set of matched quadruples in G is if-invariant.
Proof of Claim: Note that {YI,Y2,Y3,Y4} is a matched quadruple if and only if
{Yi,Yj} E 1) for alII:::; i < j:::; 4 and supp(Yj)A.suPP(Yj) = !\{SUpp(Yk): 1:::; k:S 4}
and V{SUPP(Yk): 1 :::; k :::; 4} f. 1. These conditions are clearly preserved by if, which
yields the claim.
Now take {gl, hd, {g2, h2} E 1) such that {gl, hd£.{U2, h2}. Let {gl, hd and
{g2, h 2} be matched covers. We can find {g3, h3} E 1) and {Y4, h4} E 1) with
matched covers {g3, h3} and {g4, h 4} such that

SUPp(gl, hI )USUPP(g3, h3), SUPP(g3, h3)UsuPP(g4, h4 ) and SUPP(g2, h 2)UsuPP(g4, h 4 )

are moieties, the spines l7(gj, hi) all agree and the rib cages of SUPP(gl, hd and
SUpp(g3, h3) are disjoint, as are the rib cages ofsupp(g3, h3) and SUPP(g4, h4)' and the
Automorphisms of Quotients of Symmetric Groups 245

rib cages ofsupp(g4' h4) and SUPP(g2, h2). Then each of {yl, hI, Y3, h3}, {Y3, h3, Y4, h4}
and {Y4, h4' Y2, h2} is a matched quadruple of G. Hence, by the claim, their images
under <p are all matched quadruples of G. Thus

as required. 0

It follows from the last lemma that <p has a natural induced action on the set of
almost-equality classes of w-sequences which enumerate moieties of n. We are now
in a position to construct a near-bijection fIP corresponding to <p. Let n = no u nl
be a partition of n into moieties, and let {y, h}, {k,!} E V have matched covers
{g, h}, {k, I} respectively such that u(g, h) is an enumeration of no and u( k, l) is an
enumeration of nl . Since <p preserves the Boolean algebra B, we see that if {gl, h'}
and {k', I'} are matched covers of{y<p, h<p} and {k<p,!<p} respectively, then U(g', h')
and u(kl,I'), regarded as sets, have finite intersection and their union is cofinite in
n. There is a near-bijection fIP of n which induces a map taking u(g, h) to U(gl, h')
and taking u(k, I) to U(k', I') (except on a finite set).
The next lemma shows essentially that this action preserves subsequences. (Note
that these subsequences are not necessarily increasing subsequences; that is, they
may involve permutations of the elements of the sequence, so may not induce the
same ordering.)

Lemma 4.4 Let <p E Aut(G). Suppose that {y, h}, {k, I} E V have matched covers
{g, h}, {k, I} respectively, and that u(g, h) meets supp( k) n supp( I) in a subsequence
of u(k, I), that is, there is A ~ wand an injection x: A - w such that if i E A then
the ith element of u(g, h) is the X(i)th element of u(k, I), and if i ¢ A then the ith
element of u(g, h) does not lie in u( k, I). Then there are matched covers {g', h'} and
{k', I'} of {y<p, h<p} and {k<p,!<p} respectively such that if i E A then the ith element
of U(g', h') is the X(i)th element of U(k', I'), and if i ¢ A then the ith element of
U(gl, h') does not occur in u( k', I').

Proof. By replacing (k, I) by a pair [-equivalent to it, and using Lemma 4.3, we
may suppose that

supp( k) n supp( I) n supp(g) n supp( h) = supp(g, h) n supp( k, /).

It follows that we may choose matched covers {g', h'} of {g<p, h<p} and {k', I'} of
{k<p,/<p} so that

supp( k') n supp(l/) n SUpp(g') n supp( h') = SUpp(g', h') n SUpp(k', I').
Hence there is a subset B of wand an injection X' : B - - + w such that if i E B
then the ith element of U(gl, h') is the X' (i)th element of u( k', I'), and if i ¢ B then
246 Alperin, Covington, and Macpherson

the ith element of ()(g' , h') is not in ()( k' , I'), for almost all i E w. We must show that
A = B and X = X' almost everywhere (for then, by finite adjustments of g', h', k', [I,
we may arrange that equality holds everywhere).
It is clear that ghk has cycle type C(S) where
S = {2ni - 1: i ~ A} U {2ni + nX(i) - 2: i E A} U {ni: i rt. X(A)}.
Similarly, g' h' k' almost has cycle type
S = {2ni -1: i ~ B} U {2ni + nx'(i) - 2:i E B} U {ni:i rt. X'(B)}.
It follows by Lemma 4.1(a),(b),(c) and (e) that AL1B is finite and XCi) = X'(i) for
almost all i E w, as required. 0

Lemma 4.5 Let f'P be as defined above. Then the =.-class of f'P is independent of
the choice of no, 0,1 and of g, ft, k, l.

=
Proof. Pick another partition 0, 0,;; U ni into moieties, and let {g* , it *}, {k* , [*} E
'0 have matched covers {g*, h *}, {k* , I*} so that ()(g* , h *) is an enumeration of 0,;;
and ()( k* ,1*) is an enumeration of ni. Define a corresponding near-bijection f;
of n. We must show f'P =. f;. This, however, follows almost immediately from
Lemma 4.4. 0

To complete the proof of the theorem, we must prove the following lemma.
Lemma 4.6 Let f'P be as above, and let 1'1' be the element of NS(n) corresponding
to f'P. Then for all g E C(2)inf we have 1; 1 g1'1' = g'P.

Proof. Let g := IliEw (a2ia2i+d· Choose permutations x, yES such that {x, y} E
'0, the set {x,y} is a matched C(T)-pair, and such that ()(x,y) is the sequence
(ai : i E w). Then supp(g) = supp(x)!\supp(y), so supp(g'P) = supp(x'P)!\suPP(y'P).
Also, {x'P, Y'P} E D, so we may choose a corresponding matched C(T)-pair {x', V'}.
Let ()(X', V') = (b i : i < w). Then by Lemma 4.5, ad'P = bi for all but finitely
many i E w. In particular, f:;;lgf'P and Ili<w(b2i b2i+d are =.-equivalent. Also, by
Lemma 3.6 (or a direct argument), g'P E C(2)inf, and if g' E S with g' = g'P, then
the symmetric difference SUpp(g')L1 {b i : i E w} is finite. It remains to show that g'
and TIiEw(hib2i+d differ by a finitary permutation.
Let u := gx. Then u has cycle type lW Il((n2i + n2i+d J : i < w). This says
that, for each i < w, 9 pairs up the n2i-cycle of x with the n2i+l-cycle of x. By
Lemma 3.7, 'It'P and it are conjugate in G, so have preimages in S with cycle types
differing finitely. If g' induces the transposition (bib j ) then x' g' has an (n2i + n2i+1)-
cycle. It now follows from Lemma 4.1( d) that g' differs finitely from TIiEw (b 2i b2 i+ j),
as required. 0

Proof of Theorem 1.1. The main theorem follows immediately from Lemma 4.6.
For, since the conjugacy class C(2)inf generates G, conjugation by 1'1' induces the
action of'P on all elements of G.
Automorphisms of Quotients of Symmetric Groups 247

References
[1] R. Baer, Die K ompositionsreihe der Gruppe alter eineindeutigen Abbildungen
einer unendlichen Menge auf sich, Studia Math, 5(1935),15-17.

[2] John D. Dixon, Peter M. Neumann and Simon Thomas, Subgroups of small
index in infinite symmetric groups, Bull. London Math. Soc. 18(1986),580-586.

[3] H. D. Macpherson and Peter M. Neumann, Subgroups of infinite symmetric

groups, J. London Math. Soc. (2),42(1990) 64-84.

[4] Fred Richman, Maximal subgroups of infinite symmetric groups, Canad. Math.
Bull. 10(1967), 375-381.
[5] M. Rubin, On the automorphism groups of homogeneous and saturated Boolean
algebras, Algebra Universal is 9(1979), 54-86.

[6] W. R. Scott, Group Theory, Prentice Hall, New Jersey, 1964.

[7] Stephen W. Semmes, Infinite symmetric groups, maximal subgroups, and filters,
Preliminary report, Abstracts Amer. Math. Soc., 69(1982), 38.
[8] J. K. Truss, On recovering structures from quotients of their automorphism
groups, this volume.
Other Mathematics and Its Applications titles of interest:

J.-F. Pomrnaret: Partial Differential Equations and Group Theory. New Perspec-
tives/or Applications. 1994,473 pp. ISBN 0-7923-2966-X
Kichoon Yang: Complete Minimal Surfaces 0/ Finite Total Curvature. 1994,
157 pp. ISBN 0-7923-3012-9
N.N. Tarkhanov: Complexes o/Differential Operators. 1995,414 pp.
ISBN 0-7923-3706-9
L. Tamassy and J. Szenthe (eds.): New Developments in Differential Geometry.
1996, 444 pp. ISBN 0-7923-3822-7
w.e. Holland (ed.): Ordered Groups and Infinite Permutation Groups. 1996,
255 pp. ISBN 0-7923-3853-7