7

8
5


r
e
v
i
s
i
o
n
:
2
0
0
2
-
0
1
-
1
8







m
o
d
i
f
i
e
d
:
2
0
0
2
-
0
1
-
1
9


ALMOST-FREE E-RINGS OF CARDINALITY
1
R

UDIGER G

OBEL, SAHARON SHELAH, AND LUTZ STR

UNGMANN
Abstract. An E-ring is a unital ring R such that every endomorphism of the un-
derlying abelian group R
+
is multiplication by some ring-element. The existence of
almost-free E-rings of cardinality greater than 2
0
is undecidable in ZFC. While they
exist in Goedels universe, they do not exist in other models of set theory. For a regular
cardinal
1
2
0
we construct E-rings of cardinality in ZFC which have
1
-free
additive structure. For =
1
we therefore obtain the existence of almost-free E-rings
of cardinality
1
in ZFC.
1. Introduction
Recall that a unital ring R is an E-ring if the evaluation map : End
Z
(R
+
) R
given by (1) is a bijection. Thus every endomorphism of the abelian group R
+
is
multiplication by some element r R. E-rings were introduced by Schultz [20] and easy
examples are subrings of the rationals Q or pure subrings of the ring of p-adic integers.
Schultz characterized E-rings of nite rank and the books by Feigelstock [9, 10] and an
article [18] survey the results obtained in the eighties, see also [8, 19]. In a natural way
the notion of E-rings extends to modules by calling a left R-module M an E(R)-module
or just E-module if Hom
Z
(R, M) = Hom
R
(R, M) holds, see [1]. It turned out that a
unital ring R is an E-ring if and only if it is an E-module.
E-rings and E-modules have played an important role in the theory of torsion-free
abelian groups of nite rank. For example Niedzwecki and Reid [17] proved that a
torsion-free abelian group G of nite rank is cyclically projective over its endomorphism
ring if and only if G = RA, where R is an E-ring and A is an E(R)-module. Moreover,
Casacuberta and Rodrguez [2] noticed the role of E-rings in homotopy theory.
It can be easily seen that every E-ring has to be commutative and hence can not be
free as an abelian group except when R = Z. But it was proved in [6] and extended in
[5, 4], using a Black Box argument from [3], that there exist arbitrarily large E-rings
R which are
1
-free as abelian groups, which means that every countable subgroup of
R
+
is free. The smallest candidate in [6, 5, 4] has size 2

0
. This implies the existence
of
1
-free E-rings of cardinality
1
under the assumption of the continuum hypothesis.
Moreover, it was shown in [16] that there exist almost-free E-rings for any regular not
weakly compact cardinal >
0
assuming diamond, a prediction principle which holds
2000 Mathematics Subject Classication. 20K20, 20K30, 13B10, 13B25
Key words and phrases: E-rings, almost-free modules
Publication 785 in the second authors list of publication. The third author was supported by a MIN-
ERVA fellowship.
1
7
8
5


r
e
v
i
s
i
o
n
:
2
0
0
2
-
0
1
-
1
8







m
o
d
i
f
i
e
d
:
2
0
0
2
-
0
1
-
1
9


2 R

UDIGER G

OBEL, SAHARON SHELAH, AND LUTZ STR

UNGMANN
for example in G odels constructible universe. Here, a group of cardinality is called
almost-free if all its subgroups of smaller cardinality than are free.
Since the existence of
2
-free E-rings of cardinality
2
is undecidable in ordinary set
theory ZFC (see [15, Theorem 5.1] and [16]) it is hopeless to conjecture that there exist
almost-free E-rings of cardinality in ZFC for cardinals larger than 2

0
. However, we
will prove in this paper that there are
1
-free E-rings in ZFC of cardinality for every
regular cardinal
1
2

0
. Thus the existence of almost-free E-rings of size
1
in
ZFC follows.
The construction of
1
-free E-rings R in ZFC is much easier if [R[ = 2

0
, this because in
case [R[ =
1
we are closer to freeness, a property which tries to prevent endomorphisms
to be scalar multiplication. Thus we need more algebraic arguments and will utilize a
combinatorial prediction principle similar to the one used by the rst two authors in
[14] for constructing almost-free groups of cardinality
1
with prescribed endomorphism
rings.
The general method for such constructions is very natural and it will be explained in
full detail in Shelah [21, Chapter VII, Section 5]. Our notations are standard and for
unexplained notions we refer to [11, 12, 13] for abelian group theory and to [7] for
set-theory. All groups under consideration are abelian.
2. Topology, trees and a forest
In this section we explain the underlying geometry of our construction which was used
also in [14], see there for further details.
Let F be a xed countable principal ideal domain with 1 ,= 0 with a xed innite set
S = s
n
: n of pair-wise coprime elements, that is s
n
F + s
m
F = F for all n ,= m.
For brevity we will say that F is a p-domain, which certainly can not be a eld. We
choose a sequence of elements
q
0
= 1 and q
n+1
= s
n
q
n
for all n (2.1)
in F, hence the descending chain q
n
F (n ) of principal ideals satises

n
q
n
F = 0
and generates the Hausdor S-topology on F. Thus F is a dense and S-pure subring of
its S-adic completion

F satisfying q
n
F = q
n

F F for all n .
Now let T =
>
2 denote the tree of all nite branches : n 2 (n ). Moreover,

2 = Br (T) denotes all innite branches : 2 and clearly

n
T for all
Br (T) (n ). If ,= Br (T) then
br (, ) = inf n : (n) ,= (n)
denotes the branch point of and . If C then we collect the subtree
T
C
= T : if e l()C then (e) = 0
of T where l() = n denotes the length of the nite branch : n 2.
Similarly,
Br (T
C
) = Br (T) : if e C then (e) = 0
7
8
5


r
e
v
i
s
i
o
n
:
2
0
0
2
-
0
1
-
1
8







m
o
d
i
f
i
e
d
:
2
0
0
2
-
0
1
-
1
9


ALMOST-FREE E-RINGS OF CARDINALITY 1 3
and hence
n
T
C
for all Br (T
C
) (n ).
Now we collect some trees to build a forest. Let
1
2

0
be a regular cardinal and
choose a family C = C

: < of pair-wise almost disjoint innite subsets of

. Let T = v : v T be a disjoint copy of the tree T and let T

= T
C

for < . For simplicity we denote the elements of T

by instead of since it
will always be clear from the context to which the nite branch refers to. By [14,
Observation 2.1] we may assume that each tree T

is perfect for < , i.e. if n then

there is at most one nite branch
n
such that
(n+1)
,=
(n+1)
for some T

.
We build a forest by letting
T

=
_
<
T

.
Now we dene our base algebra as B

= F[z

: T

] which is a pure and dense

subalgebra of its S-adic completion

B

taken in the S-topology on B

.
For later use we state the following denition which allows us to view the algebra B

as a module generated over F by monomials in the variables z

( T

).
Denition 2.1. Let X be a set of commuting variables and R an F-algebra. If Y R
then M(Y ) will denote the set of all products of elements from Y , the Y -monomials.
Then any map X R extends to a unique epimorphism : F[X] F[(X)]. Thus
any r F[(X)] can be expressed by a polynomial
r
F[X], which is a preimage under
: There are l
1
, , l
n
in (X) such that
r =
r
(l
1
, , l
n
) =

mM({l
1
,...,ln}
f
m
m with f
m
F
becomes a polynomial-like expression.
In particular, if Z

= z

: T

( < ) and Z

= z

: T

, then as always
the polynomial ring B

can be viewed as a free F-module over the basis of monomial,

we have B

zM(Z

)
zF and a subring B

zM(Z)
zF.
Since
1
2

0
= [Br (T
C
)[ we can choose a family V

Br (T
C
) : < of
subsets V

of Br (T
C
) with [V

[ = for < . Note that for ,= < the innite

branches from V

and V

branch at almost disjoint sets since C

is nite, thus
the pairs V

, V

are disjoint. Moreover, we may assume that for any m , pairs of

branches in V

branch above m.
3. The Construction
Following [14] we use the
Denition 3.1. Let x

B

be any element in the completion of the base algebra B

.
Moreover, let V

for < . Then we dene the following branch like elements for
n which are of the form y
nx
:=

in
q
i
qn
(z

i
) +x

in
q
i
qn
(i).
7
8
5


r
e
v
i
s
i
o
n
:
2
0
0
2
-
0
1
-
1
8







m
o
d
i
f
i
e
d
:
2
0
0
2
-
0
1
-
1
9


4 R

UDIGER G

OBEL, SAHARON SHELAH, AND LUTZ STR

UNGMANN
Note that each element y
nx
connects an innite branch Br (T
C
) with nite
branches from the disjoint tree T

. Furthermore, the element y

nx
encodes the innite
branch into an element of

B

. We have a rst observation which describes this as an

equation and which is crucial for the rest of this paper.
y
nx
= s
n+1
y
(n+1)x
+z
n
+x(n) for all < , V

. (3.1)
Proof. We calculate the dierence
q
n
y
nx
q
n+1
y
(n+1)x
=

in
q
i
(z

i
) +x

in
q
i
(i)

in+1
q
i
(z

i
) x

in+1
q
i
(i) = q
n
z
n
+q
n
x(n).
Dividing by q
n
yields y
nx
= s
n+1
y
(n+1)x
+z
n
+x(n).
The elements of the polynomial ring B

are unique nite sums of monomials in Z

with coecients in F. Thus, by S-adic topology, any 0 ,= g

B

can be expressed
uniquely as a sum
g =

z[g]
g
z
,
where z runs over an at most countable subset [g] M(Z

) of monomials and 0 ,= g
z

z

F. We put [g] = if g = 0. Thus any g

B

has a unique support [g] M(Z

),
and support extends naturally to subsets of

B

by taking unions of the support of its

elements. It follows that
[y
no
] = z

: j , j n
for any V

, n and [z] = z for any z M(Z

).
Support can be used to dene the norm of elements. If X M(Z

) then
[[X[[ = inf < : X
_
<
M(Z

)
is the norm of X. If the inmum is taken over an unbounded subset of , we write
[[X[[ = . However, since cf () > , the norm of an element g B

is [[g[[ = [[[g][[ <

which is an ordinal < hence either discrete or conal to . Norms extend naturally
to subsets of B

. In particular [[y
no
[[ = + 1 for any V

.
We are ready to dene the nal F-algebra R as a F-subalgebra of the completion of
B

. Therefore choose a transnite sequence b

( < ) which runs times through the

non-zero pure elements
7
8
5


r
e
v
i
s
i
o
n
:
2
0
0
2
-
0
1
-
1
8







m
o
d
i
f
i
e
d
:
2
0
0
2
-
0
1
-
1
9


ALMOST-FREE E-RINGS OF CARDINALITY 1 5
b =

mM
m B

with nite M M(T

). (3.2)
We call these bs special pure elements which have the property that B

/Fb is a free
F-module.
Denition 3.2. Let F be a p-domain and let B

:= F[z

: T

] be the polynomial ring

over Z

as above. Then we dene the following smooth ascending chain of F-subalgebras

of

B

.
(1) R
0
= 0; R
1
= F;
(2) R

<
R

, for a limit ordinal;

(3) R
+1
= R

[y
nx
, z

: V

, T

, n ];
(4) R = R

<
R

.
We let x

= b

if b

with [[b

[[ and x

= 0 otherwise.
For the rest of this paper purication is F-purication and properties like freeness,
linear dependence or rank are taken with respect to F. First we prove some properties
of the rings R

( ). It is easy to see that R

= F[y
nx

, z

: V

, T

, n
, < ] is not a polynomial rings: the set y
nx
, z

: V

, T

, n , < is
not algebraically independent over F. Nevertheless we have the following
Lemma 3.3. For any xed n and < the set y
nx
, z

: V

, T

is
algebraically independent over R

. Thus R

[y
nx
, z

: V

, T

] is a polynomial
ring.
Proof. Assume that the set of monomials M (y
nx
, z

: V

, T

) is linearly de-
pendent over R

for some < and n . Then there exists a non-trivial linear

combination of the form
(3.3)

yY

zEy
g
y,z
yz = 0
with g
y,z
R

and nite sets Y M (y

nx
: V

) and E
y
M (Z

). We have
chosen V

= for all ,= and M(Z

) R

= . Moreover |R

| < |R
+1
| and
hence there exists a basal element z
y
B

(high enough in an innite branch) for any

1 ,= y Y with the following properties
(i) z
y
, E
y
for all y Y ;
(ii) z
y
, [ y] for all y ,= y Y ;
7
8
5


r
e
v
i
s
i
o
n
:
2
0
0
2
-
0
1
-
1
8







m
o
d
i
f
i
e
d
:
2
0
0
2
-
0
1
-
1
9


6 R

UDIGER G

OBEL, SAHARON SHELAH, AND LUTZ STR

UNGMANN
(iii) z
y
, [g
y,z
] for all y Y, z E
y
;
(iv) z
y
[y].
Now we restrict the equation (3.3) to the basal element z
y
and obtain g
y,z
z
y
z = 0 for
all z E
y
. Since z
y
, [g
y,z
] we derive g
y,z
= 0 for all 1 ,= y Y and z E
y
. Therefore
equation (3.3) reduces to

zE
1
g
1,z
z = 0. We apply M(Z

) R

= once more. Since

each z is a basal element from the set M(Z

) we get that g
1,z
= 0 for all z E
1
. Hence
g
y,z
= 0 for all y Y, z E
y
, contradicting the assumption that (3.3) is a non-trivial
linear combination.
The following lemma shows that the F-algebras R

/s
n+1
R

are also polynomial rings

over F/s
n+1
F for every n < . For < and n we can choose a set U
n
V

such that for any V

there is

U
n
with br (,

) > n and if ,

U
n
, then
br (,

) n. Obviously [U
n
[ 2
n
. Moreover, let T

= T

n
: U
n
.
Lemma 3.4. If n < , then the set X

n+1
= y
nx

, y
(n+1)x

, z

: U
n
, T

, <
is of algebraically independent over F/s
n+1
F and generates the algebra R

/s
n+1
R

.
Thus R

/s
n+1
R

= F/s
n+1
F[X

n+1
] is a polynomial ring.
Remark. Here we identify the elements in X

n+1
R

with their canonical images

modulo s
n+1
R

.
Proof. First we show that X

n+1
is algebraically independent over F/s
n+1
F. Suppose
(3.4)

yY

zEy
f
y,z
yz 0 mod s
n+1
R
with f
y,z
F and nite sets Y M
_
y
nx

, y
(n+1)x

: U
n
, <
_
and E
y

M
_

<
T

_
.
Choose a basal element z
y
[y] for any 1 ,= y Y which is a product of basal element
z

with l() = n and z

y
/ [y

] for any y ,= y

Y and moreover require z

y
/ E
y
for all
y

Y . This is possible by the choice of U

n
and T

. Restricting (3.4) to z
y
yields

zEy
f
y,z
z
y
z 0 mod s
n+1
R
hence f
yz
0 mod s
n+1
R. Therefore (3.4) reduces to

zE
1
f
1,z
z 0 mod s
n+1
F and thus
also f
1,z
0 mod s
n+1
F is immediate. This shows that the set X

n+1
is algebraically
independent over F/s
n+1
F.
Finally we must show that R

/s
n+1
R

= (F/s
n+1
F) [X

n+1
]. We will show by induction
on < that
(R

+s
n+1
R

) /s
n+1
R

(F/s
n+1
F) [X

n+1
].
7
8
5


r
e
v
i
s
i
o
n
:
2
0
0
2
-
0
1
-
1
8







m
o
d
i
f
i
e
d
:
2
0
0
2
-
0
1
-
1
9


ALMOST-FREE E-RINGS OF CARDINALITY 1 7
If = 0 or = 1 then the claim is trivial, hence assume that > 1 and for all <
we have
(R

+s
n+1
R

) /s
n+1
R

(F/s
n+1
F) [X

n+1
].
If is a limit ordinal, then (R

+s
n+1
R

) /s
n+1
R

(F/s
n+1
F) [X

n+1
] is immediate.
Thus assume that = +1. By assumption and x

we know that (x

+s
n+1
R

)
(F/s
n+1
F) [X

n+1
]. Hence equation (3.1) shows that the missing elements z
n
+s
n+1
R

( U
n
) are in (F/s
n+1
F) [X

n+1
].
For V

we can choose

U
n
such that br (,

) > n. Then using (3.1) we

obtain y
nx

nx

0 mod s
n+1
R and therefore y
nx

+ s
n+1
R (F/s
n+1
F)[X

n+1
].
By induction on m < using again (3.1) it is now easy to verify y
mx

+ s
n+1
R

(F/s
n+1
F) [X

n+1
] for every m < , U
n
and hence R

+s
n+1
R

(F/s
n+1
F) [X

n+1
]
which nishes the proof.
Now we are able to prove that the members R

of the chain R

: < are F-pure

submodules of R and that R is an
1
-free domain.
Lemma 3.5. R is a commutative F-algebra without zero-divisors and R

as an Fmodule
is pure in R for all < .
Proof. By denition each R

is a commutative F-algebra and hence R is commutative.

To show that R has no zero-divisors it is enough to show that each member R

of the
chain R

: < is an F-algebra without zero-divisors. Since F is a domain we can

assume, by induction, that R

has no zero-divisors for all < and some 1 < < . If

is a limit ordinal then it is immediate that R

has no zero-divisors. Hence = + 1

is a successor ordinal and R

is a domain. If g, h R

with gh = 0 ,= g, then we must

show that h = 0. Write g in the form
(g) g =

yYg

zEg,y
g
y,z
yz
with 0 ,= g
y,z
R

and nite sets E

g,y
M (Z

) and Y
g
M
_
y
nx
: V

_
for some
n . By (3.1) and x

we may assume n is xed. Similarly, we write

(h) h =

yY
h

zE
h,y
h
y,z
yz
with h
y,z
R

and nite sets Y

h
M
_
y
nx
: V

_
and E
h,y
M (Z

).
Next we want h
y,z
= 0 for all y Y
h
, z E
h,y
. The proof follows by induction on the
7
8
5


r
e
v
i
s
i
o
n
:
2
0
0
2
-
0
1
-
1
8







m
o
d
i
f
i
e
d
:
2
0
0
2
-
0
1
-
1
9


8 R

UDIGER G

OBEL, SAHARON SHELAH, AND LUTZ STR

UNGMANN
number of h
y,z
s. If h = h
w,z
wz

, then
gh =

yYg,zEg,y
g
y,z
h
w,z
yzwz

and from Lemma 3.3 follows g

y,z
h
w,z
= 0 for all y Y
g
, z E
g,y
. Since R

has no
zero-divisors we obtain h
w,z
= 0 and thus h = 0. Now assume that k + 1 coecients
h
y,z
,= 0 appear in (h). We x an arbitrary coecient h
w,z
and write h = h
w,z
wz

+ h

so that wz

does not appear in the representation of h

. Therefore the product gh is of

the form
(gh) gh =

yYg

zEg,y
g
y,z
h
w,z
yzwz

+gh

.
If the monomial wz

appears in the representation of (g) then the monomial w

2
(z

)
2
appears in the representation of (gh) only once with coecient g
w,z
h
w,z
. Using Lemma
3.3 and the hypothesis that R

has no zero-divisors we get h

w,z
= 0.
If the monomial wz

does not appear in the representation of (g) then g

y,z
h
w,z
= 0 for
all appearing coecients g
y,z
is immediate by Lemma 3.3. Thus h
w,z
= 0 and h = h

follows. By induction hypothesis also h = 0 and R has no zero-divisors.

It remains to show that R

is a pure F-submodule of R for < . Let g R R

such that fg R

for some 0 ,= f F and choose < minimal with g R

. Then
> and it is immediate that = + 1 for some , hence fg R

. Now
we can write
(g) g =

yYg

zEg,y
g
y,z
yz
with g
y,z
R

and nite sets Y

g
M
_
y
vkx
: v V

_
for some xed k and E
g

M (Z

) and clearly
fg =

yYg

zEg,y
fg
y,z
yz R

.
Hence there exists g

such that
fg g

yYg

zEg,y
fg
y,z
yz g

= 0.
From Lemma 3.3 follows fg
y,z
= 0 for all 1 ,= y Y
g
, 1 ,= z E
g,y
, thus g
y,z
= 0 because
R is a torsion-free F-module. Hence (g) reduces to the summand with y = z = 1, but
g = g
1,1
R

contradicts the minimality of . Thus g R

and R

is pure in R.
7
8
5


r
e
v
i
s
i
o
n
:
2
0
0
2
-
0
1
-
1
8







m
o
d
i
f
i
e
d
:
2
0
0
2
-
0
1
-
1
9


ALMOST-FREE E-RINGS OF CARDINALITY 1 9
From the next theorem follows for = 0 that R is an
1
-free F-module. We say that
R is polynomial
1
-free if every countable F-submodule of R can be embedded into a
polynomial subring over F of R. Clearly, polynomial
1
-freeness implies
1
-freeness.
Theorem 3.6. If F is a p-domain and R =

<
R

is the F-algebra constructed above,

then R is a domain of size with R/R

is polynomial
1
-free for all < .
Proof. [R[ = is immediate by construction and R is a domain by Lemma 3.5. It
remains to show that R is an polynomial
1
-free ring. Therefore let U R be a
countable pure submodule of R. There exist elements u
i
R such that
U = u
1
, ..., u
n
, ...)

R.
Here the sux denotes purication as an F-module. Let U
n
:= u
1
, , u
n
)

for
n . Hence there is a minimal
n
< such that u
i
R
n
for i n and n , which
obviously is a successor ordinal
n
=
n
+ 1. Moreover, U
n
R
n
since R
n
is pure in
R and by induction we may assume that R
n
is polynomial
1
-free. Fix n . Using
R
n
= R
n+1
= R
n
[y
mxn
, z

: V
n
, T
n
, m ] from Denition 3.2 we can
write
u
i
=

yY
i

zE
i,y
g
y,z,i
yz
with g
y,z,i
R
n
and nite sets Y
i
M
_
y
mxn
: V
n
_
for some xed m and
E
i,y
M (Z
n
). Choose the pure submodule R
Un
:= g
y,z,i
: y Y
i
, z E
i,y
, 1 i n)

R
n
of R
n
and let
U

n
:= y, z : y Y
i
, z E
i,y
, 1 i n .
By induction there is a polynomial subring L
n
R
n
of R
n
which contains R
Un
purely. Again by induction we may assume that L
n+1
is a polynomial ring over L
n
for
all n . Hence U

n
:= L
n
[U

n
]

R
n
is a polynomial ring by Lemma 3.3 and purity
of R
Un
in R
n
. Thus U
n

R
n
. By construction L
n+1
_
U

n+1

is a polynomial
ring over L
n
[U

n
] and thus the union U

n
U

n
is a polynomial ring containing U.
Similarly arguments show that R/R

is polynomial
1
-free for every < .
4. Main Theorem
In this section we will prove that the F-algebra R from Denition 3.2 is an E(F)-
algebra, hence every F-endomorphism of R viewed as an F-module is multiplication by
some element r from R. Every endomorphism of R is uniquely determined by its action
on B

which is an S-dense submodule of R. It is therefore enough to show that a given

7
8
5


r
e
v
i
s
i
o
n
:
2
0
0
2
-
0
1
-
1
8







m
o
d
i
f
i
e
d
:
2
0
0
2
-
0
1
-
1
9


10 R

UDIGER G

OBEL, SAHARON SHELAH, AND LUTZ STR

UNGMANN
endomorphism of R acts as multiplication by some r R when restricted to B

. It
is our rst aim to show that such acts as multiplication on each special pure element
x

for < . Therefore we need the following

Denition 4.1. A set W is closed if
x

W
:= F[y
nx

, z

: V

, T

, W, < , n ]
for every W. Moreover let R
W
:= F[y
nx

, z

: V

, T

, W, n ].
We have a rst lemma.
Lemma 4.2. Any nite subset of is a subset of a nite and closed subset of .
Proof. If ,= W is nite then let = max(W). We prove the claim by induction
on . If = 0, then W = 0, R
W
= F, x
0
= 0 and there is nothing to prove. If > 0,
then x

= F[y
nx

, z

: n , V

, T

, < ] and there exists a nite set

Q such that
x

F[y
nx

, z

: n , V

, T

, Q].
If Q
1
= Q (W) then max(Q
1
) < . Thus by induction there exists a closed and
nite Q
2
containing Q
1
. It is now easy to see that W

= Q
2
is as required.
Closed and nite subsets W of give rise to nice presentations of elements in R
W
.
Lemma 4.3. Let W be a closed and nite subset of and r R
W
. Then there exists
m
r

N such that r F[y

nx

, z

: V

, T

, W] for every n m
r

.
Proof. We apply induction on [W[. If [W[ = 0, then R
W
= R

= F and Lemma 4.3

holds. If [W[ > 0 then = max(W) is dened. It is easy to see that W

= W is
still closed and nite. Thus x

R
W
for all W. By induction there is m

such that
x

F[y
nx

, z

: V

, T

, W

] for every n m

( W). Any r R
W
can
be written as a polynomial
r = (y

r,l
k
r,l
x

r,l
, z

r,j
:
r,l
V

r,l
,
r,j
T

r,j
, l < l
r
, j < j
r
)
for some l
r
, j
r
N,
r,l
,
r,j
W and
r,l
V

r,l
,
r,j
T

r,j
. Let m
r

= max(m

, k
r,l
:
l < l
r
, W). Using (3.1) now it follows easily that r F[y

nx

, z

: V

,
T

, W] for every n m
r

.
We are ready to show that every endomorphism of R acts as multiplication on each
of the special pure elements x

.
7
8
5


r
e
v
i
s
i
o
n
:
2
0
0
2
-
0
1
-
1
8







m
o
d
i
f
i
e
d
:
2
0
0
2
-
0
1
-
1
9


ALMOST-FREE E-RINGS OF CARDINALITY 1 11
Denition 4.4. If R

is as above, then let G

= y
nx
, z

: V

, T

, n )
F
be
the F-submodule of R

for any < .

From (3.1) we note that x

and our claim will follow if we can show that every

homomorphism from G

to R
+
maps x

to a multiple of itself.
Proposition 4.5. If h : G

R is an F-homomorphism, then h(x

) x

R.
Proof. Let h : G

R be an F-homomorphism and assume towards contradiction that

h(x

) , x

R. For a subset V V

of cardinality we dene the F-submodule

G
V
= x

, y
nx
: V, n )

and note that z

n
: V, n G
V
from x

G
V
and (3.1). Also G
V

H =: G
V
: V V

, [V [ = ,= and we can choose

= min : G
V

H and h(G
V
) R

. There is G
V
H such that h(G
V
) R

.
We rst claim that

< and assume towards contradiction that

= and we can
choose inductively a minimal countable subset U =: U
V
V such that
(4.1) ( V )(n )( U
V
) such that
n
=
n
.
For each V we dene the countable set Y

= y
nx
: n < . Using cf () =
>
0
we can nd a successor ordinal < such that h(x

) R

and h(Y

) R

for all U. If n

and V choose n

< n and
n
U by (4.1) such that

n
=
n

n
. From Denition 3.1 and (2.1) we see that
_

_
y
nx
y
nnx
=

in
q
i
qn
(z

i
) +x

in
q
i
qn
(i)

in
q
i
qn
(z
n
i
) x

in
q
i
qn

n
(i)
=

in+1
q
i
qn
(z

i
) +x

in
q
i
qn
(i)

in+1
q
i
qn
(z
n
i
) x

in
q
i
qn

n
(i)
(4.2)
is divisible by s
n1
. Thus s
n1
divides h(y
nx
y
nnx
) for n

< n < . From

h(y
nnx
) R

and the choice of

n
U it follows that h(y
nx
) + R

R/R

is
divisible by innitely many s
n
. Hence h(y
nx
) R

since R/R

is
1
-free by Lemma
3.6. However n

was chosen arbitrarily, we therefore have h(Y

) R

for all V and

h(G
V
) R

follows, which contradicts the minimality of

. Therefore

,= .
Since h(G
V
) R

we can write h(y

ox
) =

(y

,l
m
,l
x

,l
, z

,k
: l < l

, k < k

)
for every V and suitable
,l
,
,k
<

,
,l
V

,l
and
,k
T

,k
. Recall that
polynomials

depend on V . For notational simplicity we shall assume that all

pairs (
,l
,
,k
are distinct. By a pigeon hole argument we may assume without loss of
7
8
5


r
e
v
i
s
i
o
n
:
2
0
0
2
-
0
1
-
1
8







m
o
d
i
f
i
e
d
:
2
0
0
2
-
0
1
-
1
9


12 R

UDIGER G

OBEL, SAHARON SHELAH, AND LUTZ STR

UNGMANN
generality that l

= l

and k

= k

for some xed l

, k

N for all V . Moreover,

since F is countable, we may assume that the polynomials

are independent of and

thus we can write

= . Hence
h(y
ox
) = (y

,l
m
,l
x

,l
, z

,k
: l < l

, k < k

).
We put W

=
,l
,
,k
: l < l

, k < k

, which is a nite subset of for every V .

By Lemma 4.2 we may assume that W

is closed. Moreover, possibly enlarging W

,
we also may assume that h(x

) R
W
for all V . Since

< and is regular

the ordinal

is a set of cardinality < with W

for all V . By a pigeon

hole argument there is W =
l
,
k
: l < l

, k < k

such that W

= W for all
V

for some V

V of cardinality . We rename V = V

. Let m

N such that
m

> l(
,k
) for all V and k < k

. Again, passing to an equipotent subset (of) V we

may assume by a pigeon hole argument that m

= m
1
is xed for all V . Now we
apply Lemma 4.3 to obtain h(y
ox
) F[y
nx

, z

: V

, T

, W] for V
and some n

N. By another pigeon hole argument we may assume that n

= n

does
not depend on V anymore. If m

= maxn

, m
1
we nd new presentations
(4.3) h(y
ox
) = (y

,l
mx

l
, z

,k
: l < l

, k < k

)
for every V and
l
,
k
W,
,l
V

l
and
,k
T

k
. Moreover, l(
,k
) m

for
all V and k < k

. The reader may notice that when obtaining equation (4.3) the
polynomial and the natural number k

may become dependent on again but a pigeon

hole argument allows us to unify them again and for notational reasons we stick to and
k

. Another pigeon hole argument, using that T

is countable, allows us to assume that

,k
=
k
for all V and k < k

, hence h(y
ox
) = (y

,l
mx

l
, z

k
: l < l

, k < k

).
Finally, increasing m

(and unifying and k

again) we may assume that all

,l

m
are dierent (l < l

) and that
(4.4)
,l

m
,=
k
for all V and l < l

, k < k

. Using a pigeon hole argument and the countability of

the trees T

l
we may assume that
,l

m
does not dependent on V for all l < l

.
Thus
(4.5)
,l

m
=:
l
T

l
and
k
,=
l
for all l < l

, k < k

from (4.4). Since W is closed and h(x

) R
W
we can
nally write
h(x

) =

(y

,l
mx

l
, z

,k
: l < l

, k < k

)
7
8
5


r
e
v
i
s
i
o
n
:
2
0
0
2
-
0
1
-
1
8







m
o
d
i
f
i
e
d
:
2
0
0
2
-
0
1
-
1
9


ALMOST-FREE E-RINGS OF CARDINALITY 1 13
for every W and suitable l

, k

N,
l
,
k
W. Obviously, increasing m

once more, we may assume that

(4.6)
,l

m
,=

,l

m
and
,l

m
,=
j
for all ,

W , l < l

, l

< l

, j < l

.
Now choose any n

> m

such that
(i) n

> sup(C

) for all ,=

W ;
(ii) s
n
is relatively prime to all coecients in ;
(iii) s
n
is relatively prime to all coecients in

for all W .
Using
0
< [V [ we can choose pairs of branches
1
,
2
V with arbitrarily large
branch point br (
1
,
2
) = n + 1 n

. Let U be the innite set of all such ns. An easy

calculation using (3.1) shows
y

1
ox
y

2
ox
=
_

ln
s
l
_
(y

1
nx
y

2
nx
)
and as br (
1
,
2
) = n + 1 we obtain
(4.7) y

1
ox
y

2
ox

_

ln
s
l
_
x

mod s
n+1
R.
We now distinguish three cases.
Case 1: If br (

1
,l
,

2
,l
) > n + 1 for some l < l

then from (3.1) follows

y

1
,l
mx

l
y

2
,l
mx

l
0 mod s
n+1
R.
Case 2: If br (

1
,l
,

2
,l
) = n + 1 for some l < l

then from (3.1) follows

y

1
,l
mx

l
y

2
,l
mx

l
+s
n+1
R x

l
R +s
n+1
R.
We have chosen br (
1
,
2
) = n + 1 > n

> sup(C

) for all ,=

W .
Hence n + 1 can not be the splitting point of pairs of branches from dierent levels
and
l
. Thus
l
= and the last displayed expression becomes
y

1
,l
mx
y

2
,l
mx
+s
n+1
R x

R +s
n+1
R.
Case 3: If k = br (

1
,l
,

2
,l
) < n+1 for some l < l

then m

< k by (4.5). From (3.1)

and the choice of n we see that y

1
,l
nx
appears in some monomial of h(y

1
ox
y

2
ox
)
with coecient relatively prime to s
n+1
. By an easy support argument (restricting to
7
8
5


r
e
v
i
s
i
o
n
:
2
0
0
2
-
0
1
-
1
8







m
o
d
i
f
i
e
d
:
2
0
0
2
-
0
1
-
1
9


14 R

UDIGER G

OBEL, SAHARON SHELAH, AND LUTZ STR

UNGMANN

1
,l

k
and using (4.4), (4.5) and (4.6)) this monomial can not appear in h(x

). From
Lemma 3.4 now follows
h(y

1
ox
y

2
ox
)
_

ln
s
l
_
h(x

) , 0 mod s
n+1
R
which contradicts (4.7).
Therefore, for all n U we obtain
_

ln
s
l
_
h(x

) s
n+1
R +x

R.
The elements

ln
s
l
and s
n+1
are co-prime, thus
h(x

nU
s
n+1
R +x

R.
Using that U is innite, we claim

nU
s
n
R +x

R = x

R,
which then implies h(x

) x

R and nishes the proof.

The special pure elements are of the form (3.2) thus x

mM
m for some nite
subset M of M(T

). Choose y

nU
s
n
R + x

R. Then there are f

n
, r
n
R for n U
such that
(4.8) y s
n
f
n
= x

r
n
.
Put R

= [x

], y, f
n
, r
n
: n U)

and let L be the pure polynomial subring of R that

contains R

and exists by Theorem 3.6. Hence equation (4.8) holds in L. We may

assume that the nite support M of x

is contained in a basis of L and hence the

quotient L/x

L is free and therefore S-reduced. This contradicts

(4.9) y s
n
f
n
mod x

L
which follows from equation (4.8) for every n U unless y x

L and hence y x

R.

We are now ready to prove that R is an E(F)-algebra.

Main Theorem 4.6. Let F be a countable principal ideal domain with 1 ,= 0 and
innitely many pair-wise coprime elements. If
1
2

0
is a regular cardinal, then
the F-algebra R in Denition 3.2 is an
1
-free E(F)-algebra of cardinality .
7
8
5


r
e
v
i
s
i
o
n
:
2
0
0
2
-
0
1
-
1
8







m
o
d
i
f
i
e
d
:
2
0
0
2
-
0
1
-
1
9


ALMOST-FREE E-RINGS OF CARDINALITY 1 15
Proof. If h is a F-endomorphism of R viewed as F-module, then we must show that h
is scalar multiplication by some element b R. From Proposition 4.5 for h G

there
exists an element b

R such that h(x

) = x

for any < , where the x

s run
through all special pure elements.
Now let U

be a countable subset of V

for every < as in (4.1). Then

R

= F[y
nx

, z

: U

, T

, < , n ]
is a countable subalgebra of R

. Since is regular uncountable there exists for every

< an ordinal

< such that h(R

) R

. We put C = < : ( < )(

<
) which a cub in . Intersecting with the cub of all limit ordinals we may assume
that C consists of limit ordinals only. If C, then similar arguments as in the proof
of Proposition 4.5 after equation (4.1), using the fact that R/R

is
1
-free show that
h(R

) R

for every < and taking unions h(R

) R

.
Let us assume for the moment that there is some

C such that for every special

pure element r B

we have b
r
R

. Suppose r
1
and r
2
are two distinct pure
elements with b
r
1
,= b
r
2
. Then choose

< C such that r

1
, r
2
R

and T

with
, ([r
1
] [r
2
]). Then
(4.10) b

+b
r
1
r
1
= h() +h(r
1
) = h( +r
1
) = b
+r
1
( +r
1
) = b
+r
1
+b
+r
1
r
1
.
Now note that R

is an R

-module and that R/R

is torsion-free as an R

-module.
Moreover, b

, b
r
1
and b
+r
1
are elements of R

, hence is not in the support of either

of them. Thus restricting equation (4.10) to we obtain
b

= b
+r
1

and therefore b

= b
+r
1
. Now equation (4.10) reduces to b
r
1
r
1
= b
+r
1
r
1
and since R
is a domain we conclude b
r
1
= b
+r
1
. Hence b
r
1
= b

and similarly b
r
2
= b

, therefore
b
r
1
= b
r
2
which contradicts our assumption. Thus b
r
= b does not depend on the special
pure elements r B

and therefore h acts as multiplication by b on the special pure

elements of B

. Thus h is scalar multiplication by b on B

and using density also on R.

It remains to prove that there is

< such that for every r B

we have b
r
R

.
Assume towards contradiction that for every C there is some element r

such
that b

= b
r

, R

. We may write r

and also b
r

as elements in some polynomial ring

over R

, hence r

=
r

(x

i
: i < i
r

) and b

=
b

( x

i
: i < i
b

). Thus
r

and
b

are polynomials over R

and the x

i
s and x

i
are independent elements over R

. By a
pigeon hole argument we may assume that for all C we have i
r

= i
r
and i
b

= i
b
for some xed i
r
, i
b
N. Now choose n < and note that canonical identication
7
8
5


r
e
v
i
s
i
o
n
:
2
0
0
2
-
0
1
-
1
8







m
o
d
i
f
i
e
d
:
2
0
0
2
-
0
1
-
1
9


16 R

UDIGER G

OBEL, SAHARON SHELAH, AND LUTZ STR

UNGMANN
:

<
R

/s
n
R

<
(R

+ s
n
R)/s
n
R is an epimorphism. Let
r

and
b

be the
images of the polynomials
r

and
b

under . Since [

<
(R

+ s
n
R)/s
n
R[ < for
every < and C consists of limit ordinals the mapping : C R/s
n
R, (
r

,
b

)
is regressive on C. Thus application of Fodors lemma shows that is constant on some
stationary subset C

of C and without loss of generality we may assume that C = C

.
For C choose
1
,
2
C such that
1
<
2
and x

i
, x

j
R

1
for all i < i
r
, j < i
b
. Let
R

be the smallest polynomial ring over R

generated by at least the elements x

1
i
, x

2
i
and x

1
i
, x

2
i
such that a
1
a
2
= a
3
and a
2
, a
3
R

implies a
1
R

. We may choose
R

= R

[H] as the polynomial ring where H RR

contains the set x

1
i
, x

2
i
, x

1
j
, x

2
j
:
i < i
r
, j < i
b
. We now consider
(4.11) b
r

+r

2
(r

+r

2
) = h(r

+r

2
) = h(r

) +h(r

2
) = b

+b

2
r

2
.
By choice of R

and r

, r

2
, b

, b

2
R

follows b
r

+r

2
R

. If some x

i
appears in the
support of b
r

+r

2
, then the product x

i
x

2
j
appears on the left side (for some j < i
b
)
of (4.11) but it does not on the right side - a contradiction. Similarly, no x

2
i
can
appear in the support of b
r

+r

2
. Thus (b
r

+r

2
b

)r

= (b
r

+r

2
b

2
)r

2
and therefore
b
r

+r

2
= b

= b

2
. Hence b

2
R

2
. But this contradicts the choice of r

2
. The
existence of

such that all elements b

r
related to special pure elements are in R

is
established.
Corollary 4.7. There exists an almost-free E-ring of cardinality
1
.
Remark 4.8. We note that the Main Theorem could also be proved for cardinals
1

2

0
which are not regular. The proof for cf () = would be much more technical
and complicated.
References
[1] R. Bowshell and P. Schultz, Unital rings whose additive endomorphisms commute, Math. Ann.
228 (1977), 197-214.
[2] C. Casacuberta, J. Rodrguez and J. Tai, Localizations of abelian Eilenberg-Mac-Lane spaces
of nite type, Prepublications, Universitat Aut` onoma de Barcelona 22 (1997).
[3] A.L.S. Corner and R. G obel, Prescribing endomorphism algebras, Proc. London Math. Soc.
50(3) (1985), 447-479.
[4] M. Dugas, Large E-modules exist, J. of Algebra 142 (1991), 405-413.
[5] M. Dugas and R. G obel, Torsion-free nilpotent groups and E-modules, Arch. Math. 45 (4)
(1990), 340-351.
[6] M. Dugas, A. Mader and C. Vinsonhaler, Large E-rings exist, J. Algebra (1) 108 (1987),
88-101.
7
8
5


r
e
v
i
s
i
o
n
:
2
0
0
2
-
0
1
-
1
8







m
o
d
i
f
i
e
d
:
2
0
0
2
-
0
1
-
1
9


ALMOST-FREE E-RINGS OF CARDINALITY 1 17
[7] P. Eklof and A. Mekler, Almost free modules, Set-theoretic methods, North-Holland, Amster-
dam 1990.
[8] T. Faticoni, Each countable reduced torsion-free commutative ring is a pure subring of an E-ring,
Comm. Algebra 15(12) (1987), 2545-2564.
[9] S. Feigelstock, Additive Groups Of Rings Vol. I, Pitman Advanced Publishing Program, Boston-
London-Melbourne (1983).
[10] S. Feigelstock, Additive Groups Of Rings Vol. II, Pitman Research Notes in Mathematics Series
169, (1988).
[11] L. Fuchs, Innite Abelian Groups - Volume I, Academic Press, New York-London (1970).
[12] L. Fuchs, Innite Abelian Groups - Volume II, Academic Press, New York-London (1973).
[13] L. Fuchs, Abelian Groups, Hungarian Academy of Science, Budapest (1958).
[14] R. G obel and S. Shelah, Indecomposable almost free modules - the local case, Canadian Journal
of Math. 50 (1998) 719 738.
[15] R. G obel and S. Shelah, On the existence of rigid
1
-free abelian groups of cardinality
1
, in
Abelian Groups and Modules Proceedings of the Padova Conference (1994), 227-237.
[16] R. G obel and L. Str ungmann, Almost-free E(R)-algebras and E(A, R)-modules, Fundamenta
Mathematicae 169 (2001), 175-192.
[17] G. Niedzwecki and J. Reid, Abelian groups cyclic and projective as modules over their endo-
morphism rings, J. Algebra 159 (1993), 139-149.
[18] R.S. Pierce and C. Vinsonhaler, Classifying E-rings, Comm. Algebra 19 (1991), 615-653.
[19] J. Reid, Abelian groups nitely generated over their endomorphism rings, Springer Lecture Notes
in Math. 874 (1981), 41-52.
[20] P. Schultz, The endomorphism ring of the additive group of a ring, J. Aust. Math. Soc. 15 (1973),
60-69.
[21] S. Shelah, book for Oxford University Press, in preparation
Fachbereich 6 Mathematik, University of Essen, 45117 Essen, Germany
E-mail address: R.Goebel@uni-essen.de
Institute of Mathematics, Hebrew University, Givat Ram, Jerusalem 91904, Israel
E-mail address: shelah@math.huji.ac.il
Institute of Mathematics, Hebrew University, Givat Ram, Jerusalem 91904, Israel
E-mail address: lutz@math.huji.ac.il