7

3
9


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
2
0







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
2
3


CONSTRUCTING SIMPLE GROUPS FOR LOCALIZATIONS
R

UDIGER G

OBEL (ESSEN) AND SAHARON SHELAH (JERUSALEM)

Abstract. A group homomorphism : A H is called a localization of A if every
homomorphism : A H can be extended uniquely to a homomorphism : H H
in the sense that = . This categorical concept, obviously not depending on the
notion of groups, extends classical localizations as known for rings and modules. More-
over this setting has interesting applications in homotopy theory, see the introduction.
For localizations : A H of (almost) commutative structures A often H resembles
properties of A, e.g. size or satisfying certain systems of equalities and non-equalities.
Perhaps the best known example is that localizations of nite abelian groups are nite
abelian groups. This is no longer the case if A is a nite (non-abelian) group. Libman
showed that A
n
SO
n1
(R) for a natural embedding of the alternating group A
n
is a localization if n is even and n 10. Answering an immediate question by Dror
Farjoun and assuming the generalized continuum hypothesis GCH we recently showed
in [12] that any non-abelian nite simple has arbitrarily large localizations. In this
paper we want to remove GCH so that the result becomes valid in ordinary set theory.
At the same time we want to generalize the statement for a larger class of As. The
new techniques exploit abelian centralizers of free (non-abelian) subgroups of H which
constitute a rigid system of cotorsion-free abelian groups. A known strong theorem on
the existence of such abelian groups turns out to be very helpful, see [5]. Like [12], this
shows (now in ZFC) that there is a proper class of distinct homotopy types which are
localizations of a given EilenbergMac Lane space K(A, 1) for many groups A. The
Main Theorem 1.3 is also used to answer a question by Philip Hall in [13].
1991 Mathematics Subject Classication. Primary 20E06, 20E32, 20E36, 20F06, 20F28, 20K40,
20K20; Secondary: 14F35.
Key words and phrases: localizations of groups, simple groups which are complete, free products
with amalgamation, HNN-extension, EilenbergMacLane spaces
The authors are supported by the project No. G 0545-173, 06/97 of the German-Israeli Foundation for
Scientic Research & Development.
GbSh:739 in Shelahs list of publications.
1
7
3
9


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
2
0







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
2
3


2 R

UDIGER G

OBEL AND SAHARON SHELAH

1. Introduction
A homomorphism : A H in some category is a localization if every homomorphism
: A H in the commutative diagram
(1.1)
A

H

H
extends uniquely to a homomorphism : H H.
Such localization functors L

A

= H with respect to derive from modules and rings,
have there a long history and are considered in many recent papers in group theory for
non-commutative cases and in connection with homotopy theory, see e.g. [2, 17, 3]. It
turned out to be of special interest to investigate properties of A which carry over to
L

A - or not. Examples for groups are the properties to be commutative, nilpotent of

class at most 2, or the condition to be a ring. In particular cases the size of H relates
to the size of A, see a summary in [12]. The relation to homotopical localizations can
be looked up in [3], see also Dror Farjouns book [6]. Here we want to concentrate on
the just mentioned cardinality problem mentioned in the abstract:
If A is nite abelian, then every localization : A H is obviously epic, hence
[H[ [A[. Moreover, if A is torsion abelian then [H[ [A[

0
as shown in [17] by
Libman. In contrast to this localizations of Z are the E-rings, see [3] and by Dugas,
Mader and Vinsonhaler [8] (using [5]) there are arbitrarily large E-rings. The question
about the size of L

A for nite, non abelian groups A still remains. As also mentioned in

the abstract, Libman [18] has shown that for particular alternating groups A = A
n
there
are localizations L

A of size 2

0
. Moreover assuming GCH any nite non abelian simple
group A has arbitrarily large localizations, as recently shown in [12]. From our new main
result we will see that GCH can be removed. Using stronger algebraic arguments, like
abelian centralizers of free (non-abelian) groups and the existence of large rigid families
of cotorsion-free abelian groups, we are able to avoid the old combinatorial setting (the
Hart Laamme Shelah game from [15]), hence GCH. As in [12] we will use the following
denition.
Denition 1.1. Let A ,= 1 be any group with trivial center and view A Aut(A) as
inner automorphisms of A. Then A is called suitable if the following conditions hold:
(1) A is a nite group.
(2) If A

Aut(A) and A

= A then A

= A.
(3) Aut(A) is complete.
Note that Aut(A) has trivial center because A has trivial center. Hence the last
condition only requires that Aut(A) has no outer automorphisms. It also follows from
this that any automorphism of A extends to an inner automorphism of Aut(A). A group
A is complete if A has trivial center zA and any automorphism is inner. If h A then
we denote by
h

: A A (x xh

= h
1
xh) the function which conjugation by h.
7
3
9


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
2
0







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
2
3


CONSTRUCTING SIMPLE GROUPS FOR LOCALIZATIONS 3
We also recall the easy observation from [12] which is a consequence of the classication
of nite simple groups:
All nite simple groups are suitable.
Also note that there are many well-known examples of suitable groups which are not
simple.
If is a cardinal, then
+
is the successor cardinal of . A partial homomorphism
between two groups is a homomorphism between subgroups accordingly. Moreover, if
U G is a subgroup of G, then the centralizer of U in G is the subgroup
c
G
U = h G : [h, U] = 1,
where [h, U] = [h, u] : u U) is the subgroup generated by the commutators [h, u] =
h
1
u
1
hu.
Denition 1.2. If / is a family of groups and G is any group, then G[/] denotes the
/-socle which is the subgroup of G generated by all copies of A / in G. If / = A,
we write G[A].
Then we have the following
Main Theorem 1.3. Let / be a family of suitable groups and be an innite cardinal
such that

0
= . Then we can nd a group H of cardinality =
+
such that the
following holds.
(1) H is simple. Moreover, if 1 ,= g H, then any element of H is a product of at
most four conjugates of g.
(2) Any A / is a subgroup of H and two dierent groups in / have only 1 in
common when considered as subgroups of H. If / is not empty, then H[/] = H.
(3) Any monomorphism : A H for some A / is induced by some h H, that
is there is some h H such that = h

A.
(4) If A

H is an isomorphic copy of some A /, then the centralizer c

H
A

= 1
is trivial.
(5) Any monomorphism H H is an inner automorphism.
Note that the second property of (2) follows from the rst property of (2) together
with (1). Also (5) can be virtually strengthened replacing monomorphism by nontrivial
homomorphism, which is also due to (1). The group theoretical techniques derive from
standard combinatorial group theory and can be found in the book by Lyndon and
Schupp [19]. We will also use a theorem concerning the existence of complicated abelian
groups from [5]. For clarity the proof will be restricted to the case when / is a singleton.
The extension to arbitrary sets / is easy and left to the reader. The reader may also
ponder about our hypothesis that all members of / are nite. In fact it turns out that
there are many innite groups A such that / = A can not be extended to H as in
the Theorem 1.3, see [14].
7
3
9


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
2
0







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
2
3


4 R

UDIGER G

OBEL AND SAHARON SHELAH

We are now ready to answer Dror Farjouns question in ordinary set theory ZFC.
Corollary 1.4. Any nite simple group has localizations of arbitrarily large cardinality.
The localization A H induces a map between EilenbergMac Lane spaces
K(A, 1) K(H, 1)
which turns out to be a localization in the homotopy category; [18]. Hence these exam-
ples show the following
Corollary 1.5. Let A be a nite simple group. Then K(A, 1) has localizations with
arbitrarily large fundamental group.
A discussion of these corollaries is given in [12], they easily derive from the Main
Theorem 1.3, see also [12]. The Main Theorem 1.3 will also be used to answer a problem
of Philip Hall from 1966 in [13] mentioned in the Kourovka notebook. There is a class
of groups G such that any extension of G by a copy of G is isomorphic to G. Only some
of the properties of the groups in our Main Theorem 1.3 will only be used for the Hall
problem.
2. Free Products With Amalgam and HNN-Extensions
The following lemma was shown in [12, Lemma 2.1]. It was basic for the proof of the
main theorem of [12] and it will be used here again. The non-trivial proof needs that A
is nite.
Lemma 2.1. Let H = G
1

G
0
G
2
be the free product of G
1
and G
2
amalgamating a
common subgroup G
0
= G
1
G
2
. If A is a nite subgroup of H, then there exist
i 1, 2 and y H such that A
y
G
i
.
Hence we have a
Corollary 2.2. Let G be any group, and : G
0
G
1
be an isomorphism between two
subgroups of G. Consider the HNN-extension H = G, t : t
1
ht = (h), h G
0
). If A
is a nite subgroup of H, then there exists a y H such that A
y
is contained in G.
We want to rene the well-known notion malnormality and say
Denition 2.3. If is a cardinal and L G are groups, then L is -malnormal in G
if
[L L
g
[ < for all g G L.
This is used in the following
Lemma 2.4. Let L G be groups, K = U L be a direct product and H = G
L
K be
a free product over L. Suppose that
7
3
9


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
2
0







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
2
3


CONSTRUCTING SIMPLE GROUPS FOR LOCALIZATIONS 5
(i) L is -malnormal in G,
(ii) h H G is an element such that [G G
h
[, and
(iii) if e L and [c
L
(e)[, then e = 1.
Then the following holds.
(a) There are 1 ,= y U, x, z G such that h = xyz.
(b) If [c
G
(h)[ then x = z
1
and c
G
(h) = L
z
.
(c) G G
h
L
z
.
Proof. We distinguish two cases depending on the position of h.
Case 1: Let h K = U L. Then we can write h = xy = xyz with x L, y U
and z = 1. If y = 1, then h = x L G contradicting (ii), hence y ,= 1 and (a) follows.
Suppose c c
G
(h) L, h K G and recall K G = K L then h = c
1
hc is reduced
of length 3 and of length 1 in H = G
L
K, a contradiction, hence
c
G
(h) = c
L
(h).
We have h = xy from above. Since [y, L] = 1, c
L
(h) = c
L
(xy) = c
L
(x) follows. If x = 1
then c
L
(h) = L, x = z = 1 and (b) holds in this case. If x ,= 1 then by (iii) follows
[c
L
(x)[ = [c
L
(h)[ = [c
G
(h)[ < and (b) holds trivially.
If g G G
h
, then g = h
1
fh for some f G, hence h = f
1
hg. Note that
h K L. If g, f G L then h = f
1
hg has length 1 and 3, a contradiction. If
g G L, f L (respectively f G L, g L) then h = (f
1
h)g has length 1 and
2, which is impossible. If f, g L, then h = xy and xy = f
1
xyg = f
1
xgy. Thus
g = x
1
fx = f
x
and G G
h
L = L
z
.
Case 2: If h H K, then let h = b
1
b
n
be in reduced form for H = G
L
K,
hence 1 < n and alternately b
i
is an element of G L and K L. Let X
i
be the element
of G L, K L with b
i
X
i
and let X

i
be the other element of G L, K L. If
b
i
K we surely may assume that b
i
U as the L-part of b
i
can be absorbed into the
amalgam L. If x G G
h
then x = h
1
yh G for some y G, hence hx = yh and if
w
1
= b
1
b
n
w
2
= y
1
b
1
b
n
x, x, y G and w
1
= w
2
(2.2)
then we claim that
x, y X
1
= X
n
, and 3 n is odd. (2.3)
We distinguish various cases:
(1) If x X

n
, y X

1
then w
2
is in reduced form and has length n+2 and l(w
1
) = n
contradicts (2.2)
7
3
9


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
2
0







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
2
3


6 R

UDIGER G

OBEL AND SAHARON SHELAH

(2) If x X
n
, y X

1
, b
n
x / L, (y
1
b
1
)b
2
b
n1
(b
n
x) is reduced of length n +1. So
l(w
1
) = n and (2.2) is impossible.
(3) The dual case x X

n
, y X
1
, y
1
b
1
/ L is similar to (2).
(4) If x X

n
, y X
1
and y
1
b
1
L then w
2
= (y
1
b
1
b
2
) b
n
x and w
1
are both
in reduced form of length n but y
1
b
1
b
2
X
2
and from w
1
= w
2
follows b
1
X
2
hence
b
1
X
1
is a contradiction.
(5) The dual case x X
n
, y X

1
and b
n
x L is similar.
(6) If x X

n
, y L then w
2
= (y
1
b
1
)b
2
b
n
x has length n and l(w
1
) = n but
x X

n
and b
n
X
n
is impossible for (2.2).
(7) the dual case x L, y X

1
is similar.
Finally we have the case
(8) x X
1
L, y X
n
L, hence b
1
b
n
= (y
1
b
1
)b
2
b
n1
(b
n
x) and both
sides are reduced of length n. By uniqueness we nd t
1
, . . . , t
n1
L such that
b
1
t
1
= y
1
b
1
, t
1
1
b
2
t
2
= b
2
, t
1
2
b
3
t
3
= b
3
, . . . , t
1
n1
b
n
= b
n
x.
From x, y G follows X
1
= X
n
and n is odd. We noted that n ,= 1, hence 3 n and
the claim (2.3) is shown.
Note that t
i
= t
iy
depends on y in (2.2) and the last displayed equations give us
t
1y
= (y
1
)
b
1
, t
2y
= t
b
2
1y
, . . . , x = (t
1
n1,y
)
bn
.
We consider the pairs (y
1
, b
1
), (t
1y
, b
2
), (t
2y
, b
3
), . . . of the last equalities. In the rst
pair the rst element may not be in L, in the second pair the second element may not
be in G, but the third pair has both these properties. If 5 n then the third pair exists
and t
3y
L, the equation above shows that
t
b
3
2y
= t
3y
L
b
3
L for all y G G
h
.
Hence
[t
3y
L L
b
3
: y G[,
by assumption (ii) of the lemma, so [L L
b
3
[. Condition (i) of the lemma implies
b
3
L, but this contradicts the reduced form of w
1
= b
1
b
n
. Hence n = 3 and
h = b
1
b
2
b
3
and from the last claim b
1
, b
3
G hence b
2
K so, as mentioned above,
without less of generality b
2
U, and if we let x = b
1
, y = b
2
and z = b
3
then (a) of the
lemma holds.
Now it is easy to show that (b) and (c) hold:
(b) We may assume (a) and that we are not in Case 1, hence h = xyz H K with
y U and x, z G L. The element h = xyz is in reduced normal form.
7
3
9


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
2
0







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
2
3


CONSTRUCTING SIMPLE GROUPS FOR LOCALIZATIONS 7
If c c
G
(h), then h = c
1
hc and we have
xyz = (c
1
x)y(zc) both sides in reduced normal form.
By uniqueness there are t
1
, t
2
L such that
xt
1
= c
1
x, t
1
1
yt
2
= y, t
1
2
z = zcx.
From y U, t
2
L, K = U L follows [y, t
2
] = 1, hence t := t
1
= t
2
and the last
displayed equations become
xt = c
1
x, t
1
z = zc.
Hence c = (t
1
)
x
1
= (t
1
)
z
and c
G
(h) L
x
1
L
z
equivalently c
G
(h)
x
L L
zx
. If
zx G L, then [c
G
(h)[ < by (i), and (b) holds trivially.
If zx = l L then c
G
(h) L
z
L
x
1
= L
z
, the element h becomes h = xyz =
x(yl)x
1
= z
1
(ly)z and [y, l] = 1.
From c
G
(h) L
z
, zx = l and h = z
1
(ly)z follows L
z
c
G
(z
1
(ly)z) = c
G
(ly)
z
or equivalently c
G
(ly) L. Hence c
G
(ly) = c
L
(ly) = c
L
(l) by [L, y] = 1. However
[c
L
(ly)[ = [c
L
(l)[ and (iii) implies l = 1. We derive h = z
1
yz and c
G
(y) L from
above. Obviously L c
G
(y), so c
G
(h) = c
(
y
z
) = L
z
and (b) follows.
(c) If g G G
h
, then g
1
= h
1
ch for some c G, hence h = chg and from (a) we
have h = xyz. We get that
xyz = (cx)y(zg) and both sides in reduced normal form of length 3.
Again there are t
1
, t
2
L with xt
1
= cx, t
1
1
yt
2
= y, t
1
2
z = zg and y U. As before
t = t
1
= t
2
L and hence t L, xt = cx, t
1
z = zg. We get g = z
1
t
1
z L
z
and (c)
is also shown.
We must extend -malnormal to sets of subgroups, as in the
Denition 2.5. A set L of subgroups is -disjoint in G if each L L has size [L[ =
and [L
g
L

[ < for all L ,= L

L and g G.
Iterating Lemma 2.4 we get a
Lemma 2.6. Let L = L
1
, . . . , L
n
be a nite collection of subgroups of G such that
(a) Each group in L is -malnormal in G.
(b) L is -disjoint in G.
If 0 m n, K
i
= U
i
L
i
, M
i
= K
i

L
i
G (i m) and
H
0
= G, H
m
=
G
M
i
: i m for m ,= 0,
then the following holds for m n.
7
3
9


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
2
0







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
2
3


8 R

UDIGER G

OBEL AND SAHARON SHELAH

(i) Each L
i
is -malnormal in H
m
for m < i n.
(ii) L is -disjoint in H
m
.
(iii) If h H
m
G and [c
G
(h)[, then there are g G, 1 l m, r U
l
with
h = r
g
.
(iv) If h H
m
G and [G G
h
[, then there is 1 l m such that h M
l
.
Proof. The proof is by induction on m. If m = 0, then (i), . . . , (iv) hold by hypothesis.
Suppose (i), . . . , (iv) holds for m. From M
m+1
= K
m+1

L
m+1
G follows
H
m+1
= H
m

G
M
m+1
= H
m

L
m+1
K
m+1
= H
m

L
m+1
(L
m+1
U
m+1
). (2.4)
(i) If h H
m
L
k
and m + 2 k n, then (i) holds by induction hypothesis. Hence
we also may assume that h H
m+1
H
m
and suppose for contradiction that
[L
k
L
h
k
[. (2.5)
The assumptions of Lemma 2.4 hold, hence we may apply (a) of the lemma and can
express h = xyz with x, z H
m
L
m+1
and 1 ,= y U
m+1
. From (2.5) and Lemma 2.4
(c) follows L
k
L
h
k
L
z
m+1
, hence L
k
L
h
k
L
z
m+1
L
k
. From k ,= m + 1, (2.5) and
hypothesis (b) we get the contradiction
[L
k
L
h
k
[ [L
k
L
z
m+1
[ < .
(ii) From (2.4) we have a canonical projection : H
m+1
H
m
with ker = U
m+1
.
If 1 i ,= j n and h H
m+1
such that L
h
i
L
j
H
m+1
has size at least , then also
[L
h
i
L
j
[. But h H
m
contradicts the induction hypothesis for (ii).
(iii) Let
h H
m+1
G such that [c
G
(h)[. (2.6)
If h H
m
G, then the induction hypothesis applies and (iii) follows. We may assume
that h H
m+1
H
m
. By Lemma 2.4(a) we have h = xyz with x, z H
m
L
m+1
and
1 ,= y U
m+1
. From Lemma 2.4(b) follows
x = z
1
and c
G
(h) = L
z
m+1
, (2.7)
hence h = y
z
. If z G, then (iii) is shown. Otherwise z = x
1
H G. We want to
derive a contradiction, showing that this case does not happen.
By Lemma 2.4(c), (2.6) and (2.7) we have c
G
(h) G G
h
L
z
m+1
= c
G
(h), hence
c
G
(h) = G G
h
= L
z
m+1
(2.8)
Hence we have that
[L
z
m+1
[ = [G G
z
[ (2.9)
7
3
9


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
2
0







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
2
3


CONSTRUCTING SIMPLE GROUPS FOR LOCALIZATIONS 9
and by induction hypothesis from (iv) for z in place of h we nd an l m such that
z G
L
l
K
l
H
m
. Now we apply Lemma 2.4(a) to write z = abc with a, c G and
b U
l
. From (2.9) and Lemma 2.4(c) we get G
x
G L
c
l
. Using c
G
(h) G
x
G L
c
l
and (2.6) we also have c
G
(h) L
c
l
L
z
m+1
, hence = [L
c
l
L
z
m+1
[ which contradicts (a).
(iv) Let h H
m+1
G and [G G
h
[. Again, if h H
m
then (iv) follows by
induction hypothesis, hence we may assume that h H
m+1
H
m
. By Lemma 2.4 (a)
we have h = xyz with x, z H
m
L
m+1
and 1 ,= y U
m+1
. If x, z G then
h = xyz G
L
m+1
K
m+1
and by induction hypothesis also (iv) follows. We may assume
that x, z G is not the case, so without restriction let z / G. From Lemma 2.4(c)
follows
G G
h
L
z
m+1
H
m
. (2.10)
If w G G
h
then w
z
1
L
m+1
G. By hypothesis on h we derive that also
[G G
z
1
L
m+1
[ (2.11)
Now, using (2.10),(2.11) and the induction hypothesis (iv) for H
m
, we nd 1 l m
such that z G
L
l
K
l
H
m
. Using Lemma 2.4(c) for c
1
in place of h and (2.11) there
is z

such that G G
c
1
L
z

l
, hence
G G
c
1
L
m+1
L
z

L
m+1
and l ,= m+ 1.
Finally we apply (2.11) once more. By Lemma 2.4(a) we get the contradiction on
cardinals [L
z

l
L
m+1
[ < .
The last Lemma 2.6 extends to innite sets L. We have an immediate
Corollary 2.7. Let L be a collection of subgroups of G such that
(a) Each group in L is -malnormal in G.
(b) L is -disjoint in G.
If K
L
= U
L
L, M
L
= K
L

L
G and H =
G
M
L
: L L, then the following holds.
(i) Each L is -malnormal in H for L L.
(ii) L is -disjoint in H.
(iii) If h H G and [c
G
(h)[, then there are g G, L L and r U
L
with
h = r
g
.
(iv) If h H G and [G G
h
[, then there is L L such that h M
L
.
Similar to polynomials over a eld K which are elements of K[x], we will say for a
group G that
7
3
9


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
2
0







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
2
3


10 R

UDIGER G

OBEL AND SAHARON SHELAH

Denition 2.8. A word w over G in a free variable x is an element of Gx). We will
write w = w(x) and may substitute elements of an over-group.
Lemma 2.9. Let G = G
1
G
2
G
3
be a free product of groups, let w
i
(x) be words over
G
1
(1 i 3) and let x
2
G
2
, x
3
G
3
. Then the following holds.
(1) If w
1
(x
2
x
3
) = w
2
(x
2
)w
3
(x
3
), then w
1
(x
2
x
3
) = tx
2
x
3
u, w
2
(x
2
) = tx
2
t

and
w
3
(x
3
) = t

1
x
3
u for some t, u, t

G
1
.
(2) If also w
2
= w
3
, then w
1
(x) = x
u
for u G
1
.
Proof. Note that it is enough to consider G = G
1
x
2
) x
3
). Write w
2
(x
2
) = t
1
t
n
with t
i
G
1
x
2
) in normal form (from alternate factors). Similarly, write w
3
(x
3
) =
u
1
u
m
with u
i
G
1
x
3
) in normal form. Then
w := w
2
(x
2
) w
3
(x
3
) = t
1
t
n
u
1
u
m
.
If t
n
x
2
) or u
1
x
3
), then w is in normal form as well. Otherwise t
n
u
1
G
1
and w = t
1
t
n1
(t
n
u
1
)u
2
u
m
is in normal form. If also w
1
(x) = v
1
v
k
with
v
i
G
1
x) is in normal form. We also may assume that v
2
x) without loss of
generality. Then writing v
i
= x
m
i
if v
i
x), we have that
w
1
(x
2
x
3
) = v
1
(x
2
x
3
)
m
2
v
3
v
k
is in reduced normal form. Hence
w
1
(x
2
x
3
) = v
1
x
2
x
3
v
3
= w
2
(x
2
) w
3
(x
3
) = t
1
x
2
(t
3
u
1
)x
3
u
3
and it follows that t
3
u
1
= 1, v
3
= u
3
and t
1
= v
1
. We get
w
1
(x
2
x
3
) = t
1
x
2
x
3
u
3
, w
2
(x
2
) = t
1
x
2
t
3
and w
3
(x
3
) = t
1
3
x
2
u
3
.
If we put t
1
= t, t
3
= t

and u
3
= u, then (i) follows.
If also w
2
(x) = w
3
(x), then txt

= t

1
xu, hence t

= u, t = t

1
. It follows that
w
1
(x
2
x
3
) = u
1
x
2
x
3
u = (x
2
x
3
)
u
as well as w
2
(x
2
) = x
u
2
and w
3
(x
3
) = x
u
3
.
The following lemma describes centralizers of nite subgroups in free products with
amalgamation.
Lemma 2.10. Let H = G
1

G
0
G
2
be the free product of G
1
and G
2
amalgamating a
common subgroup G
0
. Let A G
1
be a non trivial nite subgroup and let x H be an
element which commutes with all elements of A. Then either x G
1
or A
g
G
0
for
some g G
1
.
We repeat the short proof from [12].
7
3
9


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
2
0







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
2
3


CONSTRUCTING SIMPLE GROUPS FOR LOCALIZATIONS 11
Proof. Suppose [x, A] = 1, x , G
1
and h A. Express x in a reduced normal form
x = g
1
g

1
g
n
g

n
,
that is, g
i
G
1
G
0
, (1 < i n) and g

i
G
2
G
0
, (1 i < n). The relation
h
1
x
1
hx = 1 yields the following
h
1
g

1
n
g
1
n
g

1
1
(g
1
1
hg
1
)g

1
g
n
g

n
= 1.
By the normal form theorem for free products with amalgamation [19, Theorem 2.6
p. 187], this is only possible if g
1
G
1
and g
1
1
hg
1
G
0
for all h A, so A
g
1
G
0
.
This concludes the proof.
By similar arguments we have
Lemma 2.11. Let G be any group, and : G
0
G
1
be an isomorphism between two
subgroups of G. Consider the HNN-extension H = G, t : t
1
ht = (h), h G
0
). If
A is a non trivial nite subgroup of H and x H such that [x, A] = 1, then x is in a
conjugate of G.
Let pInn (G) denote the set of partial inner automorphisms, which are the isomor-
phisms : G
1
G
2
where G
1
, G
2
G such that can be extended to an inner
automorphism of G. Hence pInn (G) are all restrictions of conjugations to subgroups
of G.
Denition 2.12. In addition we will use Denition 1.1.
(1) Let A

A = AutA be xed groups such that A is suitable.
(2) / consists of all groups G such that A

A G, and any isomorphic copy of A
in G has trivial centralizer in G. That is,
/ = G :

A G, if A

= A

G, x G with [A

, x] = 1, then x = 1.
We have an easy lemma from [12].
Lemma 2.13. If G and G

are in / then G G

/.
By a well-known result of Schupp [21] any automorphism is partially inner for some
group extension. We will rene this result below. If G is any group in / and is an
isomorphism between two subgroups of G isomorphic to A, we will need that is an
partially inner automorphism in some extension G H /. This follows by using
HNN-extensions as we will show next.
Lemma 2.14. Let G / and B G be a subgroup isomorphic to A. Then there is
H / such that G H and Aut(B) H.
7
3
9


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
2
0







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
2
3


12 R

UDIGER G

OBEL AND SAHARON SHELAH

Proof. Let

B = Aut(B) and N = n
G
(B) the normalizer of B in G. If

B G then
let H = G. Suppose that

B , G. Note that N = G

B, so we can consider the free
product with amalgamation H := G
N

B. We shall show that H /. Let A

H
be a subgroup isomorphic to A and 1 ,= x H such that [A

, x] = 1. By Lemma 2.1
we can suppose that A

G or A

B. Suppose that A

G, the other case is

easier. Let x = g
1
g
2
g
n
be written in a reduced normal form. First suppose that
n = 1. If x = g
1
G then x = 1 since G /, and this is a contradiction. Hence
x = g
1


B N. As in Lemma 2.10 we deduce that A

= (A

)
g
1
N, thus A

= B since
B is suitable. Hence g
1
N is a contradiction. If n = 2, then we obtain (A

)
g
1
= (A

)
g
1
2
a contradiction unless A

N, so A

= B. So both g
1
and g
2
are in N, which also is a
contradiction. Similarly, if n 3 we have g
n1
and g
n
in N. This is again impossible.
This concludes the proof.
By the previous lemma we can suppose that if B G /, and if B

= A, then

B G as well. If C, B G, A

= B

= C and

C,

B are conjugate in G then C and B
are also conjugate. Indeed, if g G such that g

:

C

B, then C
g

B is a subgroup
isomorphic to B, hence C
g
= B by Denition 1.1
Lemma 2.15. Let G / and B

B G. Suppose that A and B are isomorphic but
not conjugate in G. Let :

A

B be any isomorphism. Then the HNN-extension
H = G, t : t
1
ht = (h) for all h

A)
is also in /.
Proof. see [12, proof of Lemma 3.5].
Lemma 2.16. Let C and B be isomorphic to A and suppose C

C G / and
B

B G. If : C B is any isomorphism, then there is G H / such
that pInn (H). Moreover, H can be obtained from G by at most two successive
HNN-extensions.
Proof. see [12, proof of Lemma 3.6].
Lemma 2.17. Let G / and suppose that G

/ or G

does not contain any subgroup

isomorphic to A. Let g G and g

with o(g) = o(g

). Then (G G

)/N / where
N is the normal subgroup of G G

generated by g
1
g

G G

.
Proof. The group H = (GG

)/N is a free product with amalgamation, hence G and G

can be seen as subgroups of H respectively. Suppose that we have a subgroup A

H
isomorphic to A and x H such that [A

, x] = 1. By Lemma 2.1 we can assume that

7
3
9


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
2
0







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
2
3


CONSTRUCTING SIMPLE GROUPS FOR LOCALIZATIONS 13
A

is already contained in G. Suppose that x ,= 1. By Lemma 2.10 it follows that either

x G or a conjugate of A

is contained in g). In the rst case x = 1 from G / is a

contradiction. The second case is obviously impossible. Thus H /.
Lemma 2.18. Let H = G
G
0
G

be the free product of G and G

amalgamating a
common subgroup G
0
. If any X G, G

, G
0
is in / such that monomorphisms from
A to X are induced by inner automorphisms of X, then H / as well.
Proof. Let A

H be a subgroup isomorphic to A, and 1 ,= x H such that [A

, x] = 1.
By Lemma 2.1 we can assume that A

G
0
and x = g
1
g
2
g
n
, is written in a reduced
form of length bigger than two. Then we have g

1
: A

(A

)
g
1
both of them inside
G
0
. By the choice of G
0
there exists g G
0
such that g

: A

(A

)
g
1
. We can also
suppose that the automorphism group

A

is already in G
0
by Lemma 2.14. Hence the
composition (g
1
1
g)

: A

is an automorphism, which is inner by completeness.

Thus, g
1
1
g G
0
and g
1
G
0
. This is a contradiction, since x was written in a reduced
form.
Proposition 2.19. Let G be a group in /. Let g, f G, where o(f) = o(g) = and
g does not belong to the normal subgroup generated by f. Then there is a group H /
such that G H and g is conjugate to f in H.
Proof. Let : f) g) be the isomorphism mapping f to g. By hypothesis , pInn G.
As in Lemma 2.16 consider the HNN-extension H = G, t : t
1
ft = g). We must show
that H /. Clearly [H[ < and consider any A

with A

= A

G and any x H
with [A

, x] = 1. As above we may assume that A

G and x H with [A

, x] = 1.
Now we apply Lemma 2.10.
Recall Denition 1.2 of an A-socle G[A].
Lemma 2.20. If g G /, then there is a group H /, such that G H, with
[H[ = [G[
0
and g H[A].
Proof. Suppose that o(g) = and that g , G[A]. Let A
1
and A
2
be two isomorphic
copies of A. Choose a non trivial element h A and let h
1
and h
2
be its copies in A
1
and A
2
respectively. Now dene
H = (G A
1
A
2
)/N
where N is the normal subgroup generated by g
1
h
1
h
2
. Then H / by Lemma 2.17
and moreover g H[A].
7
3
9


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
2
0







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
2
3


14 R

UDIGER G

OBEL AND SAHARON SHELAH

If o(g) = n < we rst embed G (G K)/N where K is dened by the repre-
sentation x
1
, x
2
: (x
1
x
2
)
n
= 1) and N is the normal closure of g
1
x
1
x
2
. Then by the
Lemma 2.17 (G K)/N /. Now, since o(x
1
) = o(x
2
) = , we can apply the rst
case.
3. Construction of rigid groups
We want to use the following natural denition where we slightly abuse the notion of
a free product as customary for external and internal direct products.
Denition 3.1. If G is a group and U G, g G then g is free over U if g, U) =
U g).
In this section we want to construct from a given suitable group A as in Denition
1.1 certain rigid groups G containing A. For the rest of the paper we make the following
assumptions on the cardinals , and :
(1) is an innite regular cardinal (cf = ).
(2) =

and =
+
is the successor cardinal of .
Moreover note that =
0
will be good enough in the following.
We want to apply (in Construction 3.4) the following theorem on torsion-free abelian
groups.
Theorem 3.2. For each subset X of the set (the cardinal) there is an
1
-free
abelian group G
X
of cardinal such that the following holds.
Hom(G
X
, G
Y
) =

Z : if X Y
0 : if X , Y
Remark 3.3. A proof of the theorem can be found in Corner, G obel [5, p.465]. An
abelian group is
1
-free if all its countable subgroups are free abelian.
The next section is a short description for the construction of the group H of Main
Theorem 1.3. Let =
+
be the cardinal above and assign four disjoint stationary
subsets S
i
(i = 0, 1, 2, 3) such that each ordinal is a limit ordinal of conality
cf () = if S
0
S
1
S
2
and cf () = if S
3
. Moreover, identify the group

A as a set with a xed interval [,

0
) of ordinals in . We also will need three lists of
maps, elements and pairs of elements each with repetitions respectively. Let
L
0
= x

: S
0

and let
L
1
= h

: A A

; S
1

where h

runs through all bijective maps from A to subsets of with repetitions for
each map. Finally choose an enumeration of pairs
L
2
= (y

, z

) : S
2

7
3
9


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
2
0







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
2
3


CONSTRUCTING SIMPLE GROUPS FOR LOCALIZATIONS 15
also with repetitions for each pair. From [S
0
[ = [S
1
[ = [S
2
[ = follows that L
0
, L
1
and L
2
exist. Now we are ready to dene H. The denition is by transnite induction.
The inductive steps are also called approximations, see Shelah [23] or G obel, Rodriguez,
Shelah [12].
The set of approximations in the Construction 3.4 is just the collection P of initial
sequences p = H

: <
p
for any
p
< of the nal group H =

<
H

. More
generally, the members of p could depend on p, i.e. p = H
p

: <
p
. Then P becomes
naturally a poset by component-wise inclusion, and any unbounded sequence in P gives
rise to a group H. This obvious generalization may be useful for other constructions, in
this case it can also be applied for nding a family of 2

non-isomorphic groups like H.

Again, for transparency we will restrict to only one group H and the ordering on P is
just extending the initial sequence p by some members H

satisfying the
Construction 3.4. We dene an ascending chain of subgroups H

( < ) with uni-

verse a subset of ( + 1) of cardinality whose union is H =

<
H

. The chain is
constructed by transnite induction subject to the following conditions.
(i) ( = 0) Let H
0
= H

Z

A for all be the free product of innite
cyclic groups Z and

A. Hence A

A H
0
are prescribed subgroups of any
H

.
(ii) If (S
0
S
1
S
2
S
3
), then let H
+1
= H

Z.
(iii) If S
0
and x

[A], then let H

+1
= H

Z. Otherwise apply free

products with amalgamation H
+1
= H

A A/N as in Lemma 2.20 to get that

x

H
+1
[A].
(iv) If S
1
and h

: A A

is a partial inner automorphism mapping A to

some subgroup A

of H

or h

is not an isomorphism between A and A

, then we
also put H
+1
= H

Z. Otherwise choose an HNN-extension H

+1
= H

, t

)
such that t

A = h

is inner on the extended groups, see [19].

(v) If S
2
and y

, z

from L
2
are two elements of innite order in H

such that
y

is not a conjugate of z

, then choose an HNN-extension H

+1
= H

, t

) such
that y

= z
t

. Otherwise let H
+1
= H

Z.
(vi) If S
3
, then we apply the Black Box 5.1 (i), (ii), (iii) in order to dene a
family F

= F
j
: j of free subgroups of rank of H

: There are branches

j
: (j )
given by the traps and models (H

j
,
j
,
j
) which are triples of subgroups H

j

H

, a unary function : H

j
and a partial two place function
: H

j
H

j
such that Im

j
H

j
.
7
3
9


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
2
0







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
2
3


16 R

UDIGER G

OBEL AND SAHARON SHELAH

We will say that is useful (for and ) if we can choose for any j a
strictly increasing, continuous sequence

j
: such that the following holds.
(1) (

j
(

j
()) <

j
(

j
( + 1)) and
(2) (

j
(

j
(2)),

j
(

j
(2 + 1))) <

j
(

j
(2 + 2)).
In this case is a total map and we dene x

j
() by
(

j
(

j
(2)),

j
(

j
(2 + 1))) = x

j
().
Let F
j
= x

j
() : < ) and F

= F
j
: j .
If this is not possible, we say that is useless and pick F

trivially from
branches as in the rst case but regardless of what and do. In Lemma 3.6
we will show that F

meets all requirements, in particular that each F

j
is free
of rank . Now we dene H
+1
in two steps:
Take a rigid family U
j
(j ) of torsion-free abelian groups of cardinal from
Theorem 3.2 such that
Hom(U
i
, U
j
) =
ij
Z
and let
K
j
= U
j
F
j
and M
j
= H

F
j
K
j
.
In the second step choose
H
+1
=
H
M
j
: j
be the free product with amalgamated subgroup H

. Hence H

H
+1
by the
normal form theorem, see [19, p. 187, Theorem 2.6].
(vi) Finally let H =

.
It remains to show that H meets the requirements of the Main Theorem 1.3. The
proof of condition (vi), which is based on the Black Box 5.1, will be postponed to the
next section, however all prerequisites will be established now using the following
Remarks and Notations 3.5. If S
3
and j from the Construction 3.4 (vi),
then let
j
() =

j
(

j
()), hence x

j
() H

j
(+1)
H

j
()
is free over H

j
() and the
elements x

j
() ( < ) freely generate F
j
F

. Moreover sup
<

j
() = and H

<
H

j
()
for each j . If i ,= j < , then [
j
() : <
i
() : < [ < .
First we show the
Lemma 3.6. Let H =

be as in the Construction 3.4.

(a) The groups F
j
(j ) dened in (vi) for S
3
are freely generated by the sets
x

j
() : < .
7
3
9


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
2
0







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
2
3


CONSTRUCTING SIMPLE GROUPS FOR LOCALIZATIONS 17
(b) Each F
j
is -malnormal in H

and F

is -disjoint.
(c) If y, z H and z ,= 1, then y is a product of at most four conjugates of z.
(d) Any monomorphism A H is induced by an inner automorphism of H.
(e) H = H[A].
(f) If S

denotes the innite symmetric group acting on countably many elements,

then Hom(H, S

) = 0.
Proof. (a) Comparing with Remark and Notations 3.5 we see that each x

j
()
H

j
(+1)
H

j
()
is free over H

j
()
, hence H

j
()
x

j
()) H

j
(+1)
H

. An easy
induction shows that F
j
H

is freely generated by the set x

j
() : < .
(b) If g H

F
j
for some j < , then g H

j
()
for some minimal

< from
H

<
H

j
()
. If

< , then clearly by freeness - as shown next -

x

j
() : < ) x

j
() : < )
g
= x

j
() : <

) x

j
() : <

)
g
which is a set of cardinality less then as [

[ < , and the rst part of (b) follows.

The proof of the displayed equation is by induction on < . If =

or is a
limit ordinal the assertion obviously holds. So suppose

< < is not a limit and let

U = H

j
()
, x

j
() : < ). Hence
g H

j
()
, x

j
() :

< ) = U x

j
()) H

.
If w is an element of the left hand side of the displayed equality, then there are also
two words w
1
(x), w
2
(x) free over U such that w = w
1
(x

j
()) = g
1
w
1
2
(x

j
())g. Hence
g = w
2
(x

j
())gw
1
(x

j
()) has length 1 in U x

j
()). Write w
2
= a
1
x
t
1
a
2
x
t
n1
a
n
and
w
1
= f
1
x
s
1
f
2
x
s
m1
f
m
in normal form, hence
g = w
2
gw
1
= a
1
x
t
1
a
2
x
t
n1
(a
n
gf
1
)x
s
1
x
s
m1
f
m
has length 1 which is only possible if t
1
= s
1
= 0, so w = w
1
(x

j
()) U and the claim
follows by the induction hypothesis.
Next we consider g H

and i ,= j < . We must show that [F

i
F
g
j
[ < . If
w F
i
F
g
j
, we can choose < such that g H

and Im

i
Im

j
. Then
x

i
() : < H

= x

i
() : <
1
and x

j
() : < H

= x

j
() : <
2

for some
1
,
2
< . As before we have F
i
F
g
j
H

= F
i
F
g
j
H

for all with

< < . Hence F
i
F
g
j
H

= F
i
F
g
j
H

, which has cardinality < , and (b)

is shown.
(c) If y and z have innite order in H, then (y, z) = (y

, z

) for some S
2
and it
is y = z
t
in H
+1
, so (c) follows in this case. If y has nite order, then we can write
y = y

with both y

, y

of innite order and it remains to show that y

is product of
7
3
9


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
2
0







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
2
3


18 R

UDIGER G

OBEL AND SAHARON SHELAH

at most two conjugates of z. If z has innite order, this is clear from above. If z has
nite order, then we can nd suitable elements x
i
H such that w = z
x
1
z
x
2
has innite
order. By the rst case y

= w
t
for some t H, hence y

= z
x
1
t
z
x
2
t
is product of two
conjugates and y is product of four.
(d) This is taken care of by the construction at stage (iv) for S
3
.
(e) If g H, there is S
0
such that g = g

, hence g H
+1
[A] by construction
and H[A] = H follows.
(f) If S

is the innite symmetric group acting on countably many elements, then

[S

[ = 2

0
= <
+
= . Hence [S

[ < and (f) follows because H is simple

by (c).
Corollary 3.7. H is simple and there is an element in H such that each other element
is a product of at most four of its conjugates.
Lemma 3.8. Let H be as in the Construction 3.4.
(a) If S
3
and < < , j , then F
j
is -malnormal in H

and F

is
-disjoint in H

.
(b) If A

H is an isomorphic copy of A, then c

H
A

= 1.
Proof. (a) follows from Lemma 3.6(b). (b) is based an the denition of / and also follows
by induction on for all A

( < ), using Lemma 2.15, Lemma 2.17 , Lemma

2.18 and Lemma 2.20.
We now have an implication which follows from Corollary 2.4, a
Lemma 3.9. If and H

from the Construction 3.4, < and y H

with a large centralizer [c

H(y)
[, then S
3
and there are j < , g H

and
x K
j
such that y = x
g
.
Proof. Let be minimal such that y H

. The proof is now induction on .

Clearly < and is not a limit ordinal, hence = + 1 for some . We write
D
y
= c
H
(y) and similarly C
y
= c
H
(y). From follows C
y
D
y
. For the rst part
of the lemma is enough to show that [D
y
[ < if ,= or if = , S
3
. Recall that
y H
+1
H

. We must distinguish cases depending on the position of .

If S
1
, then H
+1
= H

, t) is an HNN-extension. Let y = g
0
t

1
g
1
g
n1
t
n
g
n
be
given in normal form with g
i
H

such that there is no subword t

1
g
i
t with g
i
A or
tg
i
t
1
A

H
+1
, see the Construction 3.4 and [19, p. 181]. Note that 1 n from
y , H

. Any 1 ,= x D
y
is in H

and commutes with y, hence

x
1
g
1
n
t

1
t

1
(g
1
0
xg
0
)t

1
g
1
g
n1
t
n
g
n
= 1.
7
3
9


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
2
0







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
2
3


CONSTRUCTING SIMPLE GROUPS FOR LOCALIZATIONS 19
By the normal form theorem of HNN-extensions ([19, p. 182]) either
1
= 1 and g
1
0
xg
0

A or
1
= 1 and g
1
0
xg
0
A

. By symmetry we may assume that

1
= 1, hence
g
1
0
xg
0
A for all x D
y
. We have D
g
0
y
A and [D
y
[ [A[ < as desired.
If S
2
, then H
+1
= H

, t) is another HNN-extension and the result follows as in

the last case.
If (S
0
S
1
S
2
S
3
) then H
+1
= H

t) which is similar to the rst cases

but much easier.
If S
0
then H
+1
arrives from two extensions as before which settles this case.
We nally deal with S
3
and the free product of the M
j
s, which is
H
+1
=
H
M
j
: j < .
Now apply Corollary 2.7 to nd y = x
g
as in the lemma.
4. Proof Of The Main Theorem
The crucial part of this paper is the following
Main Lemma 4.1. Any endomorphism of the group H from Construction 3.4 is an
inner automorphism of H.
Proof. If is an endomorphism of H, then is a monomorphism because H is simple.
We will write H =

as in the construction.
Constructing modules with prescribed endomorphism rings, the most important con-
dition is nding elements x of the module (say H) such that x / H

, x)

, see the
strong case in [5, p.455]. Here we will also say that an element 1 ,= x H is strong
(for ) at if x is free over H

, hence
H

, x) = H

x) (4.12)
and
x / H

, x). (4.13)
In this case we also say that is strong for . If x is free over H

(i.e. (4.12) is true),

but (4.13) does not hold, we call x weak (for ) at , and if all free elements x over H

are weak at , we call a weak ordinal for .

We will distinguish two cases:
(A) All ordinals are strong.
(B) There is a weak ordinal

< .
The case (B) is the complementary case of (A) . We rst consider case (A):
7
3
9


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
2
0







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
2
3


20 R

UDIGER G

OBEL AND SAHARON SHELAH

For each ordinal there is a strong element x

H for at .
By a back and forth argument we can choose a closed and unbounded set C and
an enumeration C =

: < such that the following holds for all < :

(1)

,
(2) H

and
1
(dom H

) H

(3) x

H
+1
H

for all < .

(4)

: < is strictly increasing and continuous.

Note that (2) is a purity condition for , saying that H H

. Condition
(3) follows by a new enumeration of the x

s and the H

s. Let U

be the set of all

ordinals

+1 such that all elements in H

+1
are weak for over . First we claim
if is strong, then the set U

is bounded in ; ( hence [U

[ ). (4.14)
If U

is unbounded and x

H
+1
H

is strong for , then choose any U

and
y

H
+1
free over H

. This is possible, because by construction often (on a stationary

set) we choose H
+1
= H

). The weak element y

tells us
y

, y

) = H

).
On the other hand the strong element x

makes
x

/ H

, x

) and x

H
+1
H

.
Hence x

H
+1
is also free over H

, and , being in U

, requires
(x

) H

, x

).
Hence there is a word w

(y) over H

with free variable y such that y

= w

(y

). As
we assume that U

is unbounded in and =
+
, also [U

[ = > [H

[. By a pigeon
hole argument we nd an equipotent subset U U

such that for U the word

w

(y) = w(y) does not depend on . We have y

= w(y

) for all U and [U[ = .

Pick any
1
<
2
in U and consider y = y

1
y

2
. Clearly y is also free over H

and

i
U implies y H

, y), hence y = w

(y) for another word w

over H

. We can
summarize
w

(y) = y = (y

1
y

2
) = y

1
y

2
= w(y

1
) w(y

2
).
Also note that we have freeness
H

1
) y

2
) H

2
+1
and normal forms
w(y

i
) H

i
) (i = 1, 2),
7
3
9


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
2
0







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
2
3


CONSTRUCTING SIMPLE GROUPS FOR LOCALIZATIONS 21
and Lemma 2.9 applies. There is g H

such that
y

i
= w(y

i
) = y
g

i
(i = 1, 2), and similarly (x

i
) = (x

i
)
g
.
Using =
1
we derive
x

= (x

y
1

) = (x

)(y

)
1
= (x

)
g
(y
g

)
1
= x
g

,
hence x

= x
g

, H

) contradicts that x

is strong for . The claim (4.14) is

shown.
By case (A) we may apply the last claim (4.14) to all < and see that all U

s are
bounded. Hence there is a new increasing, continuous sequence E =

: < C
of ordinals such that in addition U

+1, hence U

+1
. Hence (1), (2), (3), (4)
hold for

in place of

there. In particular

and x

+1
H

is a strong
element at

.
We now want to adjust the Black Box 5.1 for application in this case (A). Let us
dene two maps

( a partial map ) and

on H which makes (H,

) into an
L-model as mentioned in the Black Box 5.1.
Let

() =

( < ) and dene

(, ) if and only if

< and let

(, ) = x

be from above. Hence

is a total function and

is partial such
that

(, ) exists if and only if

() . The group H together with

is an
L-model M with universe . We want to consider L-submodels M = (H

) of
M = (H,

), hence M is a subgroup H

of H,

a total function and

(, ) is
dened if and only if

() . As E is a cub and S
3
is a stationary set, we also nd
stationary many S
3
E, hence H

is closed under and

1
. The restriction

(denoted by

again) is a total function, and H

with these restrictions of

is an
L-model M

with universe a subset of .

By (iv) of the Black Box there is some i < for such such that (H

i
,

)
(H

). Hence H

i
is a subgroup of H with Im

i
H

i
,

i
a unary total
function on H

i
,

i
a two place function on H

i
which is dened again for (, ) if
and only if

() =

< and such that

(, ) = x

+1
H

is strong for
at

.
The Black Box 5.1 predicts some j < and a strictly increasing sequence

j
() ( < )
with sup
<

j
() = ,

j
() + 2 <

j
( + 1) for all < and

j
() [

+1
) with

j
(

j
(2)),

j
(

j
(2 + 1))) = x

j
() = x

j
()
(4.15)
for all < as dened in the Construction 3.4. If F
1
= F
j
, which is freely generated
by the x

j
() ( < ), also given by the Construction 3.4, then F
1
H

follows from
7
3
9


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
2
0







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
2
3


22 R

UDIGER G

OBEL AND SAHARON SHELAH

F
1
H

and C. Recall that H

+1
=
H
M
j
: j < . If 1 ,= r U
j
, then r / H

and r / H

by the closure property (2). However

M
j
= H

F
1
K
j
= H

F
1
(F
1
U
j
) H
+1
and U
j
H

= 1
hence c
H
(r) F
1
. Clearly r U
j
implies F
1
c
H
(r), we get the important centralizer
condition
c
H
(r) = F
1
.
Next we calculate the centralizer of the image r, using H

:
F
1
= (c
H
(r)) = c
H
(r) c
H
(r). (4.16)
By Corollary 3.9 there are F
i
= F
2
F

, g H

and f K
i
such that
r = f
g
. (4.17)
From f K
i
follows r = f
g
H
+1
. We derive the invariance
H
+1
H
+1
. (4.18)
Using (4.16), (4.17) and g H

, f K
i
we get
F
1
c
H
(r) = c
H
(f
g
) = c
H
(f)
g
= F
g
2
. (4.19)
Note that by denition g = g
r
depends on r and similarly F
2
= F
2r
. We want to show
that dierent rs give the same g and F
2
:
If r
1
,= r
2
and g
r
1
, g
r
2
are as above, then by (4.19) we have F
1
F
gr
1
2r
1
F
gr
2
2r
2
, hence
= [F
2r
1
F
gr
2
g
1
r
1
2r
2
[ and by the choice of F

(which is -disjoint by Lemma 3.6) it

follows that F
2r
1
= F
2r
2
and g
r
1
= g
r
2
. We obtain that for all r U
F
1
also r U
g
F
2
,
hence U
F
1
U
g
F
2

= U
F
2
. The family U
i
: i < of abelian groups is rigid, and this
forces F
1
= F
2
and U
F
1
is conjugation by g. The claim (4.19) becomes F
1
F
g
1
,
and g H

<
H

j
()
is an element of some H

j
()
. If (, ), then x

j
()
by (4.15) is a canonical free generator of F
1
and x

j
() H

j
()+1
H

j
()
. Hence
x

j
() H

j
()+1
H

j
(+1)
and also
x

j
()
F

+1
x

j
()
: < )
g
x

j
()
: < + 1)
g
from F
1
F
1
and Lemma 3.6. From g H

j
() and x

j
()
H

j
() for all <
follows x

j
()
x

j
()
, H

j
()
). However x

j
()
is free over H

j
()
and strong for by
denition of

, which is a contradiction. Hence case (A) does not come up.

Case (B) can be derived quite easily: There is an ordinal

< such that any free

element x H over H

is weak for , hence x H

, x). So there is a word

w
x
(y) over H

with variable y such that x = w

x
(x).
7
3
9


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
2
0







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
2
3


CONSTRUCTING SIMPLE GROUPS FOR LOCALIZATIONS 23
Pick any two free elements b
1
, b
2
over H

such that b
2
is also free over H

, b
1
). There
are many bs! And choose < such that H

b
1
, b
2
H

, also let z H be free

over H

. Hence b
1
z, b
2
z and z are free over H

and we have
H

, b
i
, z) = H

b
1
) z).
For the words w
b
i
z
(y), w
z
(y) we get (b
i
z) = w
b
i
z
(b
i
z) and z = w
z
(z) and it follows
w
b
i
z
(b
i
z) = (b
i
z) = (b
i
)(z) = w
b
i
(b
i
)w
z
(z)
and Lemma 2.9(a) applies. We can write b
i
= e
i
b
i
d
i
, z = e
z
zd
z
with e
i
, d
i
.e
z
, d
z
H

and d
i
e
z
= 1 for i = 1, 2. Hence d
1
= d
2
= d and e
z
= d
1
. We have b
i
= e
i
b
i
d and
z = d
1
zd
z
. The same argument for (b
1
b
2
z) = (b
1
)(b
2
z) gives d = d
z
. Similarly
for b
1
1
b
1
2
z
1
also e
i
= d
1
, hence b
i
= b
d
i
, z = z
d
. We conclude that = d

for all
elements b H which are free over H

. Obviously H is generated by such bs, hence

= d

on H is inner, which nishes the proof of the Main Lemma.

Proof. (of the Main Theorem 1.3) From Main Lemma 4.1 follows that we only must check
conditions 1., 2., 3. of the Main Theorem 1.3. This follows however from Lemma 3.6.
5. Appendix: A Model Theoretic Version Of The Black Box
Let L be the language (of groups in our case) with a nite vocabulary of cardinality
at most and with a unary function ( ) and a partial two place function ( , ). From
Shelah [23, Chapter IV] we adopt the following prediction principle - a model theoretic
version of the old and often used Black Box form [24] which was also used and proved
in the appendix of [5], see also [11, 7, 10] for other applications.
In order to match the setting to earlier ones, we will use terms from trees. Condition
(i) below can be viewed as a tree embedding from

into a tree in of branches below

and condition (ii) just says that distinct branches of length have only a small branch
of length < in common. Condition (iii) is the earlier requirement that the image of
a tree

can be found in any submodel of the trap (see [5] for instance) here called
(

i
, M

i
: i < ) and (iv) is the prediction of a submodel of M, earlier ([5]) this was a
module or a group together with an unwanted homomorphism , so a pair (H, ). In
our application it will be a group together with a unary map and a partial two place
map on H.
Another preliminary remark seems in order:
We will predict ordinals from a stationary subset of (and submodels), hence the
following is actually a stationary Black Box as used in [10] for instance. The reader can
either nd a proof of the group theoretic version of the Black Box by slight modication
from these references, or adopt the model theoretic version, which then has the advantage
that it is applicable in many dierent algebraic situation (including the old ones) without
any further changes. Again the proof of the model theoretic version of the Black Box is
a natural and easy modication of the existing proofs; a nal reference will be [23].
7
3
9


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
2
0







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
2
3


24 R

UDIGER G

OBEL AND SAHARON SHELAH

Black Box 5.1. (in model theoretic terms) Let L be a language just mentioned and
suppose =

< =
+
as before and let S be a stationary subset of
< : cf = .
Then there is a sequence (of traps)
((

i
, M

i
) : i < ), S)
such that the following holds.
(i)

i
: (

i
()) (i < ) is an increasing, continuous sequence with
supremum (a branch):
(ii) Any two distinct branches are almost disjoint:
If i < j < , then [Im

i
Im

j
[ < .
(iii) M

i
is an L-model with a universe of cardinality which is a subset of and
Im

i
M

i
.
(iv) If M is such an L-model with universe , then there are stationary many S
with some i < such that M

i
M.
References
[1] M. Aubry, Homotopy theory and models, Based on lectures held at a DMV Seminar in Blaubeuren
by H. J. Baues, S. Halperin and J.-M. Lemaire, DMV Seminar, Vol. 24, Birkh auser Verlag, Basel,
1995.
[2] C. Casacuberta, Recent advances in unstable localization, CRM Proceedings and Lecture Notes 6,
Amer. Math. Soc., Providence, 1994, 122.
[3] C. Casacuberta, J. L. Rodrguez and J.-Y. Tai, Localizations of abelian EilenbergMac Lane spaces
of nite type, preprint 1998.
[4] C. Casacuberta, D. Scevenels and J. H. Smith, Implications of large-cardinal principles in homo-
topical localization, preprint 1998.
[5] A. L. S. Corner, R. G obel, Prescribing endomorphism algebras - a unied treatment, Proc. London
Math. Soc. (3) 50, 447 479 (1985).
[6] E. Dror Farjoun, Cellular spaces, null spaces and homotopy localization, Lecture Notes in Math.
vol. 1622, Springer-Verlag, BerlinHeidelbergNew York 1996.
[7] M. Dugas, R. G obel, On locally nite p-groups and a problem of Philip Halls, Journal of Algebra
159 (1993), 115138.
[8] M. Dugas, A. Mader and C. Vinsonhaler, Large E-rings exist , J. Algebra 108 (1987), 88101.
[9] P. C. Eklof and A. H. Mekler, Almost Free Modules. Set-theoretic Methods, North-Holland, Ams-
terdam, New York 1990.
[10] B. Franzen, R. G obel, Prescribing endomorphism algebras. The cotorsion-free case. Rend. Sem.
Mat. Padova 80 (1989), 215 241.
7
3
9


r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
2
0







m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
2
3


CONSTRUCTING SIMPLE GROUPS FOR LOCALIZATIONS 25
[11] R. G obel, A. Paras, Outer automorphism groups of metabelian groups, Journal of Pure and Applied
Algebra 149 (2000) 251266.
[12] R. G obel, J. L. Rodrguez and S. Shelah, Large localizations of nite simple groups, to appear
Crelle Journal 2001
[13] R. G obel, S. Shelah, Philip Halls Problem On Non-Abelian Splitters, submitted to Proc. Camb.
Phil. Soc. 2000
[14] R. G obel, S. Shelah, Simple extensions of groups, in preparation 2001
[15] B. Hart, C. Laamme and S. Shelah, Models with second order properties V: A general principle,
Ann. Pure Appl. Logic, 64 (1993), 169194.
[16] T. Jech, Set Theory, Academic Press, New York, 1978.
[17] A. Libman, Cardinality and nilpotency of localizations of groups and G-modules, Israel J. Math.,
to appear.
[18] A. Libman, A note on the localization of nite groups, preprint 1998.
[19] R. C. Lyndon and P. E. Schupp, Combinatorial Group Theory, Springer Ergebnisberichte 89,
BerlinHeidelbergNew York 1977.
[20] D. Robinson, A Course in the Theory of Groups, Graduate Texts in Math. vol 80, Berlin
HeidelbergNew York 1996.
[21] P. E. Schupp, A characterization of inner automorphisms, Proc. Amer. Math. Soc. 101 (1987),
226228.
[22] S. Shelah, On a problem of Kurosh, Jonsson groups and applications, S. I. Adian, W. W. Boone,
G. Higman, eds., Word Problems II, North-Holland, Amsterdam, 1980, 373394.
[23] S. Shelah, Non Structure Theory, Oxford University Press (2001) in preparation
[24] S. Shelah, A combinatorial theorem and endomorphism rings of abelian groups II, pp. 37 86, in
Abelian groups and modules, CISM Courses and Lectures, 287, Springer, Wien 1984
[25] G. W. Whitehead, Elements of Homotopy Theory, Graduate Texts in Math. vol 61, Springer-
Verlag, BerlinHeidelbergNew York, 1978.
Fachbereich 6, Mathematik und Informatik, Universit at Essen, 45117 Essen, Germany
E-mail address: R.Goebel@Uni-Essen.De
Department of Mathematics, Hebrew University, Jerusalem, Israel, and Rutgers
University, New Brunswick, NJ, U.S.A
E-mail address: Shelah@math.huji.ac.il