5

1
9


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







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


ON THE EXISTENCE OF RIGID
1
FREE ABELIAN GROUPS
OF CARDINALITY
1
R

UDIGER G

OBEL AND SAHARON SHELAH

1. Introduction
An abelian group is said to be
1
free if all its countable subgroups are free. A
crucial special case of our main result can be stated immediately.
Indecomposable
1
free abelian groups of cardinality
1
do exist.
The rst example of any
1
free group which is not free is the BaerSpecker group
Z

, which is the cartesian product of countably many copies of the group Z of

integers, known for almost sixty years; cf. Baer [1] or [14, p.94]. Assuming CH,
this group of cardinality 2
0
=
1
is an example of a nonfree abelian group of
cardinality
1
. Under the same settheoretic assumption of the continuum hypoth-
esis it can be shown that any countable ring R with free additive group can be
realized as the endomorphism ring of an
1
free abelian group G of cardinality
1
.
The chronologically earlier realization theorem of this kind uses the weak diamond
prediction principle which follows from 2
0
< 2
1
, cf. Devlin and Shelah [6] for the
weak diamond, Shelah [28] for the case End G = Z and Dugas, G obel [7] for the
case R = End G and extensions to larger cardinals. Using, what is called Shelahs
Black Box, the existence of
1
free groups G with [G[ = 2
0
also follows from Cor-
ner, G obel [5] using Dugas, G obel [7] and combinatorial ne tuning from Shelah
[29].
Without the assumption of CH, the existence of nonfree,
1
free groups of cardi-
nality
1
follows from a more general result by Grith [18], Hill [21], Eklof [11],
Mekler [24] and Shelah in Eklof [12, p.82,Theorem 8.8]. By an induction it can be
shown, that there are
n
free groups, nonfree of cardinality
n
. The nonabelian
case is due to Higman [19, 20].
By Shelahs singular compactness theorem it is known that free abelian groups
of cardinality do not exist if is singular, e.g. if =

, cf. Eklof, Mekler [13].

Hence induction breaks down and it is more complicated to show the existence of
free, nonfree abelian groups of cardinality >

. This question is investigated

in Magidor, Shelah [23] and we just refer to this paper and restrict ourselves to
cardinals 2
0
again, and we will focus on =
1
. Only very little is known
about algebraic properties of
1
free groups of cardinality
1
, see Eklof [11] and
Eklof, Mekler [13]. Shelahs construction [27] (see also [30]) of groups also mentioned
in [12, 13] which are not separable was rened in Eda [10] prove the existence of
Part of the work for this paper was carried out while the rst author visited Rutgers University.
He would like to thank the organizers of MAMLS and the Department of Mathematics for their
support.
Number of publication 519. Research was supported by the Edmund Landau Center for re-
search in Mathematical Analysis, supported by the Minerva Foundation (Germany).
1
5
1
9


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







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


2 R

UDIGER G

OBEL AND SAHARON SHELAH

an
1
free group G of cardinality
1
such that Hom (G, Z) = 0, a result derived
independently but later by Corner, G obel [5]. Counterexamples for Kaplanskys
test problems among
1
free groups of cardinality
1
are given recently in G obel,
Goldsmith [17], realizing rings modulo some large ideal, see also [16]. Moreover,

1
separable groups of cardinality
1
serving as counterexamples of Kaplanskys
test problems were constructed in [31]. These results about
1
free groups become
special cases of our quite satisfying main theorem.
Main Theorem 4.1 If R is a ring with R
+
free and [R[ < 2
0
, then there
exists an
1
free abelian group G of cardinality with End G = R.
We have identied R with endomorphisms acting on the Rmodule G by scalar
multiplication. This result has many applications. If R = Z, we derive the existence
of
1
free abelian groups of cardinality
1
, a result which was unknown.
If is any abelian semigroup, then we use Corners ring R

, implicitly discussed in
Corner, G obel [4], and constructed for particular

s in [3] with special idempotents

(expressed below), with free additive group and [R

[ = max[[,
0
. If [[ < 2
0
,
we may apply the main theorem and nd a family of
1
free abelian groups G

(
) of cardinality
1
such that for , ,
G

= G
+
and G

= G

if and only if = .
Observe that this induces all kinds of counterexamples to Kaplanskys test problems
for suitable

s. If we consider Corners ring in [2], see Fuchs [15, p.145], then it is

easy to see that R
+
is free and [R[ =
0
. The particular idempotents in R and our
main theorem provide the existence of an
1
free superdecomposable group of car-
dinality
1
, which was unknown as well. Recall that a group is superdecomposable
if any nontrivial summand decomposes into a proper direct sum.
Finally, we remark that as the reader might suspect, it is easy to replace G in
Theorem 4.1 by a rigid family of 2

such groups with only the trivial homomorphism

between distinct members. The main theorem cannot be generalized, replacing
1
by another cardinal. In Section 5 we will show that there are many models of ZFC
(e.g. assuming MA and
2
< 2
0
) in which no
2
free group of cardinality < 2
0
has endomorphism ring Z; it is even possible that all such groups are separable and
the best one can do now is a realization theorem of the form End G = RInes G
with Ines G ,= 0 an ideal containing all endomorphisms of nite rank.
This is in contrast to the result [7], that under

any countable ring R with R

+
free
is of the form R

= End G for all uncountable regular, not weakly compact cardinal
= [G[ > [R[ such that G is free. In particular, the existence of indecomposable

2
free groups of cardinality
2
or the existence of such groups with endomorphism
ring Z is undecidable.
2. The building blocks,
1
free modules with a distinguished cyclic
submodule
Let R be a ring of cardinality [R[ < 2
0
such that R
+
is a free abelian group. In
view of Pontrjagins theorem we say that an Rmodule is
1
free if any subgroup
of nite rank is contained in a free Rsubmodule.
We have the immediate application of Pontrjagins theorem [14, p.93, Theorem
19.1.].
5
1
9


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







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


ON THE EXISTENCE OF RIGID 1FREE ABELIAN GROUPS OF CARDINALITY 1 3
Observation 2.1. If M is
1
free as Rmodule and R
+
is free, then M is
1
free
as abelian group, this means all countable subgroups are free.
Remark 2.2. If U is a nitely generated submodule of an
1
free Rmodule M of
innite rank and M/U is at, then M/U is an
1
free Rmodule as well.
Proof If S/U is a subgroup of nite rank in M/U, then S

/U denotes its puri-

cation and S

is a pure subgroup of nite rank in M, hence it is contained in a free

Rsubmodule F of M. Moreover, we nd a nitely generated summand F

of the
Rmodule F with S

and F/F

is Rfree. Also F

/U is at because M/U is
at and F

/U can be nitely presented by

F

/U 0
for some nitely generated free module F

mapping onto U F

. Hence F

/U is
projective by Rotman [25, p. 90, 91], and F/U

= F/F

/U is projective.
Finally we may assume that F/U has innite rank and F/U is free by a wellknown
argument of Kaplanskys, cf. [17], for instance. Hence M/U must be an
1
free
Rmodule.
Recall that Remark 2.2 does not hold if U is not nitely generated. Consider a
free resolution of any torsionfree abelian group A which is not
1
free: 0 U
M A 0. By Remark 2.2 in particular quotients of
1
free groups modulo
pure, cyclic subgroups are
1
free again.
Next we will construct particular
1
free Rmodules A with distinguished cyclic
submodules cR.
First we will x some more notation. Let T be a family of 2
0
almost disjoint
innite subsets of an innite set of primes. At present, we choose a xed X T
with an enumeration X = p
n
: n without repetitions. Let T =
>
2 denote
the tree of all nite branches : n 2, n < , where () = n denotes the length
of the branch . The branch of length 0 is denoted by = T and we also
write = ( n 1)

(n 1). Finally

2 = Br(T) denotes all innite branches
: 2 and clearly n T for all Br(T), n .
Let be an innite cardinal 2
0
and Y Br(T) with [Y [ = and [R[ < .
Then V

will denote the vector space over the rationals Q with basis T Y . Finally
R becomes a vector space by R
Z
Q =

R and V = V

Q

R is a vector space
of dimension . We now select an Rsubmodule A V which is generated by T
together with elements

0
= ,
n+1
=
1
p
n
(
n
+ n +(n) ) V (X)
dened inductively for all Y , n . Hence
A = A
X
= A
XY
= R,
n
R : T, Y , n ) V
depends on X T and Y Br(T). The required cyclic Rsubmodule is R. We
will show that (A, R) belongs to the category of modules we are interested in,
i.e. the following Lemma holds.
Lemma 2.3. Let (A, R) be the pair of Rmodules dened above, let B = T)
and

: A A/B be the canonical homomorphism. Then we have
(a) B is a free Rmodule and A/B =

Y
(

X
Z
R) with

X Q
of characteristic : 2 with support X.
5
1
9


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







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


4 R

UDIGER G

OBEL AND SAHARON SHELAH

(b) A is an
1
free Rmodule.
(c) A/ R is an
1
free Rmodule.
Proof (a) Clearly B =

T
R and if g A, then we use (X) to nd k
and nite sets T
1
T, Y
1
Y with
g =

Y1

k
g

T1
g

for some g

, g

R. Using (X) again, we have

g

Y1

q
k
mod B
where q
k
=

k1
i=1
p
i
by the enumeration in X and
g
q
k

XR. Clearly : Y
is QRindependent and hence

X Rindependent and (a) follows.
(b) Obviously [A[ = [Y [ = . Next we show that
(*) any nite subset of A lies in a submodule U which is free and pure in A.
For any nite subset E of A we can nd some n and a nite subset Y
0
Y
such that
E U = R,
n
R : T, () < n, Y
0
).
Obviously U is freely generated by the elements ,
n
. In order to show that U is
pure in A, consider g A and m N minimal with gm = u U.
We may write
g =

Y1

k
g

T1
g

and u =

Y2

n
u

T2
u

with g

, g

, u

, u

R and k = k() minimal for each Y

1
. Since gm = u
we have Y
1
= Y
2
and
n
u

=
k
g

m for all Y
1
from (a). If k < n for some
Y
1
= Y
2
, then we can reduce Y
1
to a smaller set Y
1
by the observation

k
g

U and
k
g

m =
n
u

and g U follows by induction. We derive k n for

all Y
1
, and suppose k > n for some .
We have p
k1
[q =

k1
i=n
p
i
and minimality of m requires p
k1
does not divide
m. On the other hand g

m = qu

and p
k1
[q hence p
k1
[g

which contradicts
minimality of k = k(). We derive k = n for all and g decomposes into a Y
part g
Y
U with g
Y
m =

Y2

n
u

and a Tpart g
T
B with g
T
m =

T1
g

.
However g
T
U, hence g = g
Y
+g
T
U as well and U is pure in A, i.e. (*) holds.
Finally A is an
1
free Rmodule by the argument in Remark 2.2 and Pontrjagins
collection of a direct sum of projective modules, see Fuchs [14, p.93, Theorem 19.1.].
Now (b) and also (c) follow from (*).
Observation 2.4. If (A, R) is as above, then A and A/ R are
1
free abelian
groups with R End A, R End (A/ R) identifying r = r id for all r R.
Observation 2.4 is immediate from Observation 2.1 and Lemma 2.3, which is all we
need in Section 3.
Moreover we will require enough splitting in A which is established by the following
Proposition 2.5. Let (A, R) be as above, where A = A
X
, X ,= P T and

P = Z
P
the obvious localization at P. Then A
X
R
P
is a free R
P
module with
a basis element, where R
P
= Z
P

Z
R is the localization of R at P.
5
1
9


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







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


ON THE EXISTENCE OF RIGID 1FREE ABELIAN GROUPS OF CARDINALITY 1 5
Proof Recall that
A
X
= R,
n
R : T, Y, n ).
Moreover X P is nite by our choice of T. We nd k such that
p
n
X : n k P = . Now we claim that
T
k
: Y
is a basis of the R
P
module A
X
R
P
. Note that T and Proposition 2.5 will
follow.
The set M = T
k
: Y is clearly independent over Q
Z
R in V and hence
freely generates the R
P
submodule
U =

mM
mR
P
= F R
P
A
X
R
P
with F =

mM
mR. It remains to show U = A
X
R
P
.
The submodule F A
X
induces a natural sequence
0 F A
X
A
X
/F 0
of Rmodules, where A
X
/F is generated by
n
+ F : Y, n > k, see Lemma
2.3(a). Using (X) we derive p
n1
... p
k+1

n

k
0 mod F where the
enumeration of primes is taken in X. These primes belong to p
n
X : n k and
cannot belong to P by our choice of k. We observe that A
X
/F is a P

group in
the wellknown sense, that A
X
/F is torsion and the order of elements is a product
of primes in P

, the complement of P. On the other hand R

P
is P

divisible, hence
(A
X
/F) R
P
= 0. Using atness of R
P
the above sequence becomes
0 F R
P
A
X
R
P
(A
X
/F) R
P
0
and A
X
/F R
P
= 0 forces A
X
R
P
= F R
P
as desired.
3. Repeating the building blocks
Let R, T and [R[ < 2
0
be as in Section 2. Then we enumerate T = X
a
:
< without repetition and it is easy to nd a family T = L

: <
of innite, almost disjoint subsets L

of without repetitions. Since Br(T) =

2
and [

2[ = 2
0
, we can also nd a family Y

Br(T) : < of sets Y

of
branches with the following additional properties
(b1) [Y

[ = for all < .

(b2) Y

has branch points above every level:

If Y

and n , there are distinct branches Y

with n = n.
(b3) The length of a branch point of branches in Y

is in L

:
If ,= Y

, then ( ) L

.
We use these three families to enumerate a family of Rmodules A
XY
constructed
in Section 2 dening A

= A
XY
for all < . Moreover we denote R
X
= R

the localization of R at the primes X

from Section 2.
Inductively we dene an ascending, continuous chain of Rmodules G

( < )
with distinguished cyclic submodules c

R G

for nonlimit ordinals < . The

module we are interested in will then be the Rmodule G = G

=

<
G

. If
= 0, let G
0
=

<
e

R be free Rmodule of rank , which is also a free abelian

group of rank because R
+
is free of rank < . We will choose elements c

for nonlimit ordinals subject to the following conditions

(c1) G

/c

R is an
1
free Rmodule
5
1
9


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







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


6 R

UDIGER G

OBEL AND SAHARON SHELAH

(c2) If c G and G/cR is an
1
free Rmodule, then [ < : c = c

[ = .
The extension G
+1
will be constructed such that condition (c1) ensures that G
is
1
free and (c2) can easily be arranged by an enumeration of elements c G

with G

/cR
1
free with [[ repetitions for all < . If = 0, then for (c1) we
may choose a basic element c
0
and we do not care for (c2).
If c

are dened for all < and is a limit, then G

=

<
G

by
continuity and it remains to construct G

from c

for = + 1. From our

choice (c1) of c

we know that G

/c

R is an
1
free Rmodule. We consider a
pushout diagram. There exists a (unique) pushout Rmodule G

with the well

known pushout mapping properties [14, p.52] or [25] in case of R-modules.
c

R G

The rst row is the canonical embedding and the rst column is an embedding by
the identication c

= . By the pushout property we now may assume that

(p

) G

= A

+G

and A

= c

R
hence G

/c

R

= G

/c

R A

/ R. The construction of G is complete.

First we will discuss freeness properties of G.
Lemma 3.1. G is an
1
free Rmodule of cardinality .
Proof If G =

<1
G

as above, then we only have to show that G

is
1
free
for any which we prove by induction. Since G
0
is free we consider > 0 and
assume that all G

( < ) are
1
free. If = + 1, then G

= A

+ G

and
(p

) holds, hence
G

/c

= G

/c

R A

/ R.
The right hand side is
1
free by Lemma 2.3 and assumption on the choice of c

.
However, if G

/c

R is
1
free, then G

must be
1
free as well.
If is a limit ordinal, then any subgroup of nite rank in G

=

<
G

is a
subgroup of G

for some < and

1
freeness follows.
The following observation plays a role in our next proposition, which provides
splittings of G coming from Proposition 2.5 and is based on
R

is divisible by all primes not in X

which is nite for ,= .

Proposition 3.2. If G =

<
G

is the Rmodule above, then G

is a free
R

module for all < .

Proof If < , then (G
+1
R

)/(G

) = (G
+1
/G

) R

because R

is a at Rmodule. We also have G

+1
/G

= A

/c

R by the pushout property

(p
+1
) and (A

/c

R) R

is a free R

module by ,= and Proposition 2.5. We

derive that (G
+1
R

)/(G

) is a free R

module, hence projective and the

rest follows inductively by an obvious basis collection. Taking into account that
G
0
R

is a free R

module, the same holds for G

.
Proposition 3.3. With the notation as above we have
(a) A

is a direct summand of G
+1
R

(b) G
+1
R

is a direct summand of GR

(c) A

is a direct summand of GR

.
5
1
9


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







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


ON THE EXISTENCE OF RIGID 1FREE ABELIAN GROUPS OF CARDINALITY 1 7
Proof Obviously (c) follows from (a) and (b) and it remains to show the rst two
assertions.
(a) Observe that G

is free by Proposition 3.2 and we may write G

=
c

as R

modules by our choice of c

. The pushout property (p

+1
) gives
G
+1
R

= H

(A

) and (a) follows.

(b) Inductively we will nd an ascending, continuous chain of complements C

of
G
+1
R

in G

for +1 and C

will verify (b). If = +1, then

C

= 0 and if is a limit ordinal between +1 and and all C

( < ) are dened,

then C

=

<
C

is already dened by continuity and C

is a complement of
G
+1
R

in G

indeed, because G

( ) is continuous at as
well. It remains to dene C

for = + 1 where C

is given. We are in the case

> + 1, hence > and ,= follows. From Proposition 2.5 we see that
c

= R

is a summand of the free R

module A

and we may write

A

= c

. Obviously C

= C

is a complement of G
+1
R

in G

by the pushout property (p

+1
).
4. Proof of the Main Theorem
The main result of this paper is the following
Theorem 4.1. If R is a ring with R
+
free and [R[ < 2
0
, then there exists an

1
free abelian group G of cardinality with End G = R.
Remark: G will be the Rmodule constructed in Section 3 and we have identied
r R with r id
G
.
Proof From Lemma 3.1 we have an Rmodule G of cardinality which is
1

free as Rmodule, hence

1
free as abelian group. Moreover R End G by our
identication and we must show that
End GR does not exist.
Such a homomorphism has a unique extension : G R

G R

because
= id extends and G R

/G = (G R

)/(G R)

= R

/R being torsion
forces uniqueness.
If c

R for all < , then c

= c

for some r

R. If < is xed, we
can choose an element c G (even in G
0
) such that G/cR is an
1
free Rmodule
cR c

R is a direct sum and G/(c +c

)R is an
1
free Rmodule as well. There
exist some , < with c = c

and c +c

= c

. We have
c

+c

= c

+c

= (c

+c

) = c

= c

= c

+c

and r

= r

= r

follows. We nd a uniform r R such that c

= c

r for all
< . However, G is generated by the set c

: < , hence = r which was

excluded.
There exists < such that c

/ c

R. We also nd > such that c

and the repetition (c2) (Section 2) of the enumeration of c

s provides < <

such that c

= c

, hence
(i) c

/ c

R and c

.
However, G

is a free R

module by Proposition 3.2 and c

is a basic element
of the R

module G

; we nd a free decomposition G

= c

C.
The pushout G
+1
= G

+A

gives G
+1
R

= (A

) C and Proposition
3.3(b) provides an R

module D such that L = C D satises

5
1
9


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







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


8 R

UDIGER G

OBEL AND SAHARON SHELAH

(ii) (A

) L = GR

, G

= c

C
where C = L (G

) by the modular law.

The element c

has a unique decomposition c

= c

r + c
with r R

and c C. If c = 0, then c

= c

R by purity
of c

is a contradiction. Hence 0 ,= c C which is a free R

module with a
basis B. The element c =

b[c]
bc
b
has a unique decomposition and a Bsupport
[c] = b B : c
b
R

0 ,= .
On the other hand c C G

and cm =

[c]
bc
b
m G

C for some m ,= 0.
However G

C G

for some < , which is contained in the free R

module
G

. Since ,= , our choice of R

, R

provides an h < such that

(iii) p
j
does not divide c C for all j > h,
where the enumeration of primes is taken in X

= p
n
: n < .
If : G
+1
R

C denotes the canonical projection induced by (ii), then

(iv) 0 ,= c = c

.
Moreover, the image of any Y

viewed as A

G
+1
R

can
be expressed by
=

b[]
br

b
with r

b
R

0
with a nite subset [] of B. Abusing notation we shall call [] the Bsupport of
as well. Recall that [Y

[ = > [R

[
0
, and it is easy to nd Y

Y

, n N
and r
b
R

for all b B such that [Y

[ = and [[][ = n, r

b
= r
b
for all Y

and b B. Next we apply the Lemma to [] : Y

(cf. Jech [22, p.225])
and nd Y

Y

, E B such that [Y

[ = and [] [

] = E for all ,=

.
Since [c] B is nite, we also nd Y Y

such that [Y [ = and [] [c] E for
all Y.
From [Y [ = >
0
we nd two distinct branches ,

Y with h =

h. The
branch point j > h of ,

belongs to L

by (b3), hence p
j
X

, where j is from
the enumeration along branches. The denition branch point gives j =

j
and (j) = 1,

(j) = 0 without loss of generality. From the relations (X

) in A

(Section 2) we have p
j
[(

j
+

j+

(j) ) in A

and p
j
[(

j
+

j+

(j) ) in A

,
hence p
j
[
j

j
+(j) =
j

j
+c

in G

and therefore p
j
[(
j

j
)+c

.
However [c] = [c

] and if d =
j

j
, then d E = 0 by our choice of
,

Y with ,=

, hence d and c are linearly independent. We conclude p

j
[c in
C which contradicts (iii) and Theorem 4.1 follows.
5. A counterexample
The reader might suspect that
1
in Theorem 4.1 can be replaced by
2
for instance.
This is the case if we assume prediction principles as (which imply CH), see
Dugas, G obel [7]. However, in general it is no longer true as follows from
Theorem 5.1. Assuming Martins axiom, any
2
free group of cardinality < 2
0
is separable.
Recall that an
1
free group is separable if any pure cyclic subgroup is a summand.
Preliminaries on (MA) can be seen in Jech [22] or Eklof, Mekler [13].
Proof If G is an
2
free group of cardinality [G[ < 2
0
, 0 ,= e G pure in G and
: eZ Z taking e = 1, then we must extend to an homomorphism : G Z.
5
1
9


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







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


ON THE EXISTENCE OF RIGID 1FREE ABELIAN GROUPS OF CARDINALITY 1 9
Let P = ; : D

Z, e D

, e = 1 where D

is a pure and nitely generated

subgroup of G. Obviously [P[ < 2
0
from [G[ < 2
0
and (P, ) is partially ordered
by extensions of maps. Suppose for a moment that P satises the hypothesis for
MA and P
g
= P : g D

is dense for all g G. Then by MA there is a

compatible set F P such that F P
g
,= for all g G. So

F = is a partial
homomorphism from G to Z. Since F P
g
,= , also g dom for all g G,
hence Hom (G, Z) and extends by denition of P. Thus it remains to show
that (P, ) satises the hypothesis of MA:
In order to show that P
g
is dense in P, we consider any P and nd

P
such that g dom

. Since G is
2
free, there is D

dom such that g D

and D

is pure and nitely generated in G by Pontrjagins theorem. Recall that

dom is pure in G, hence pure in D

and D

/dom must be nitely generated

and torsionfree. We apply Gau theorem to see that D

/dom is free, hence

D

= dom C for some C D

with C

= D

/dom . Now it is easy to

extend to a homomorphism

: D

Z. Finally, we must show that (P, )

satises ccc, the countable antichain condition. Let F P be an uncountable
subset of P. We must nd two distinct elements
i
F and P such that

i
for i = 1, 2. We may assume [F[ =
1
, hence (

F
dom )

= U, the
pure subgroup of G generated (purely) by all dom has cardinality
1
and must
be free by hypothesis on G. We select a basis B of U and replace any F by

with dom

= B

) dom with a nite subset of B

of B. The argument
given above allows to extend to a homomorphism

.
Clearly, it is enough to nd two compatible elements
i
in the new F. By the
Lemma (Jech [22, p.225]) we also nd E B and F

F such that [F

[ =
1
and
dom dom

= E for all ,=

. By a pigeonhole argument we can also

nd F

such that [F

[ =
1
and E =

E for all ,

. Now it is
clear that we can extend two of these maps ,

to dom + dom

as required.
References
[1] R. Baer: Abelian groups without elements of nite order, Duke Math. J. 3 (1937),
68122.
[2] A.L.S. Corner: Every countable reduced torsionfree ring is an endomorphism ring,
Proc. London Math. Soc. (3) 13 (1963), 687710.
[3] A.L.S. Corner: Additive categories and a theorem of W.G. Leavitt, Bull. Amer. Math.
Soc. 75 (1969), 7882.
[4] A.L.S. Corner and R. G obel: Prescribing endomorphism algebras, a unied treatment,
Proc. London Math. Soc. (3) 50(1985), 447479.
[5] A.L.S. Corner and R. G obel: On the existence of an
1
free abelian group of cardi-
nality
1
with no free summands (unpublished).
[6] K. Devlin and S. Shelah: A weak version of which follows from 2

0
< 2

1
, Israel
J. Math. 29 (1978), 239247.
[7] M. Dugas and R. G obel: Every cotorsionfree ring is an endomorphism ring, Proc.
London Math. Soc (3) 45 (1982), 319336.
[8] M. Dugas and R. G obel: Every cotorsionfree algebra is an endomorphism algebra,
Math. Zeitschr. 181 (1982), 451470.
[9] M. Dugas and R. G obel: On radicals and products, Pacic J. Math. 18 (1985), 70104.
[10] K. Eda: Cardinal restrictions for preradicals, pp.277283 in Abelian Group Theory,
Contemporary Math. 87, Providence (1989).
[11] P.C. Eklof: On the existence of free abelian groups, Proc. Amer. Math. Soc. 47
(1975), 6572.
5
1
9


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







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


10 R

UDIGER G

OBEL AND SAHARON SHELAH

[12] P.C. Eklof: Set theoretic methods in homological algebra and abelian groups, Les
Presses de lUniversite de Montreal, Montreal 1980.
[13] P.C. Eklof and A.H. Mekler: Almost free modules, settheoretic methods, North
Holland, Amsterdam 1990.
[14] L. Fuchs: Innite abelian groups, Vol. I, Academic Press, New York, 1970.
[15] L. Fuchs: Innite abelian groups, Vol.II, Academic Press, New York, 1973.
[16] R. G obel: Radicals in abelian groups, pp. 77107 in Theory of radicals, Colloqu. Math.
Bolyai 61, North Holland, Amsterdam 1993.
[17] R. G obel and B. Goldsmith: The Kaplansky Test problem an approach via radicals,
Journ. Pure and Appl. Algebra 99 (1995), 331344.
[18] Phillip Griffith: nfree abelian groups, Quart. J. Math. (2) 23 (1972), 417425.
[19] G. Higman: Almost free groups, Proc. London Math. Soc. 1 (1951), 184190.
[20] G. Higman: Some countably free groups, pp. 129150 in Proceedings Group Theory,
Singapore 1991.
[21] P. Hill: New criteria for freeness in abelian groups II, Trans. Amer. Math. Soc. 196
(1974), 191201.
[22] T. Jech: Set Theory, Academic Press, New York 1978.
[23] M. Magidor and S. Shelah: When does almost free imply free? (for groups, transver-
sals, etc.), Journ. Amer. Math. Soc. 7 (4) (1994), 769830.
[24] A.H. Mekler: How to construct almost free groups, Can. J. Math. 32 (1980), 1206
1228.
[25] J. Rotman: Homological Algebra, Academic Press, New York 1979.
[26] S. Shelah: On uncountable abelian groups, Israel J. Math. 32 (1979), 311330.
[27] S. Shelah: Whitehead groups may not be free, even assuming CH. II Israel J. Math.
35 (1980), 257 285
[28] S. Shelah: On endorigid strongly
1
free abelian groups in
1
, Israel J. Math. 40
(1981), 291295.
[29] S. Shelah: A combinatorial principle and endomorphism rings. I: On pgroups, Israel
J. Math. 49 (1984), 239257.
[30] S. Shelah: A combinatorial theorem and endomorphism rings of abelian groups. II in
Abelian groups and modules CISM Lecture Notes, Vol. 287, Springer Wien-New York
[31] B. Thome:
1
separable groups and Kaplanskys test problems, Forum Math. 2 (1990),
203212.
Fachbereich 6, Universit at Essen, 45117 Essen, Germany
E-mail address: mat100@vm.hrz.uni-essen.de
Department of Mathematics, Hebrew University, Jerusalem, Israel, and Rutgers Uni-
versity, Newbrunswick, NJ U.S.A.
E-mail address: shelah@math.huji.ac.il