7

2
7


r
e
v
i
s
i
o
n
:
2
0
0
0
-
0
5
-
0
9







m
o
d
i
f
i
e
d
:
2
0
0
0
-
0
5
-
0
9


REFLEXIVE SUBGROUPS OF THE BAER-SPECKER GROUP AND
MARTINS AXIOM
R

UDIGER G

OBEL AND SAHARON SHELAH

Abstract. In two recent papers [9, 10] we answered a question raised in the
book by Eklof and Mekler [7, p. 455, Problem 12] under the set theoretical
hypothesis of

1
which holds in many models of set theory, respectively of
the special continuum hypothesis (CH). The objects are reexive modules over
countable principal ideal domains R, which are not elds. Following H. Bass
[1] an R-module G is reexive if the evaluation map : G G

is an iso-
morphism. Here G

= Hom(G, R) denotes the dual module of G. We proved

the existence of reexive R-modules G of innite rank with G

= GR, which
provide (even essentially indecomposable) counter examples to the question
[7, p. 455]. Is CH a necessary condition to nd nasty reexive modules?
In the last part of this paper we will show (assuming the existence of super-
compact cardinals) that large reexive modules always have large summands.
So at least being essentially indecomposable needs an additional set theoretic
assumption. However the assumption need not be CH as shown in the rst
part of this paper. We will use Martins axiom to nd reexive modules with
the above decomposition which are submodules of the Baer-Specker module
R

.
1. Introduction
We will derive our results for abelian groups, but it is an easy exercise to replace
the ground ring Z by any countable principal ideal domain which is not a eld.
Just notice that we could work with one prime only! For supercompact cardinals
we refer either to Jech [13] or to Kanamori [14]. If G is any abelian group then
G

= Hom(G, Z) denotes its dual group, and G is a dual if G

= D

for some
abelian group D.
Particular dual groups are the reexive groups D, see Bass [1, p. 476]. Recall
that
=
D
: D D

(d (d))
with (d) D

and
(d) : D

Z ( (d))
is the evaluation map and D is reexive if the evaluation map
D
is an isomorphism.
Recent results about reexive and dual abelian groups are discussed in [7, 9, 10].
1991 Mathematics Subject Classication. primary: 13C05, 13C10, 13C13, 20K15, 20K25,
20K30; secondary: 03E05, 03E35.
Key words and phrases. almost free modules, reexive modules,duality theory, modules with
particular monomorphism.
This work is supported by the project No. G-545-173.06/97 of the German-Israeli Foundation
for Scientic Research & Development
GbSh 727 in Shelahs list of publications.
1
7
2
7


r
e
v
i
s
i
o
n
:
2
0
0
0
-
0
5
-
0
9







m
o
d
i
f
i
e
d
:
2
0
0
0
-
0
5
-
0
9


2 R

UDIGER G

OBEL AND SAHARON SHELAH

In the third section we will show that dual groups, in particular reexive groups
may have large summands, hence cant be essentially indecomposable without any
set-theoretic restrictions.
Theorem 1.1. If is a supercompact cardinal and H is a dual group of cardinality
, then there is a direct summand H

of H with [H

[ < for any cardinal

< .
This theorem explains that we had to use CH in [9] and alternatively below we
will use Martins axiom (and possibly negation of CH).
In order to prove a result in contrast of Theorem 1.1 we use scalar products on
the Baer-Specker group P. Recall that
P = Z

is the set of all elements

x =

i
x
i
e
i
with x
i
Z
and e
i
P dened by the Kronecker symbol, addition is dened component-wise.
Throughout this paper we will adopt the convention in writing elements of P as
displayed in the last formula. The Baer-Specker group P has the subgroup S of
all elements x of nite support, that is x
i
= 0 for almost all i . The crucial
subgroup for constructing reexive groups is the Z-adic closure D of S in P. This
will be our target in Section 3. We will also show that the the endomorphism ring
of such a reexive abelian group can be Z modulo the ideal of all endomorphisms
of nite rank. We have the the following
Theorem 1.2. (ZFC + MA) There are two subgroups H
i
(i = 1, 2) of the Baer-
Specker group P with the following properties.
(i) S H
i

D are pure.
(ii) H
i
is
1
-free and slender.
(iii) There is a natural bilinear form : H
1
H
2
Z arriving from scalar
product on P which induces H

1

= H
2
and H

2

= H
1
such that H
1
, H
2
are
reexive.
(iv) H
i
Z ,

= H
i
for i = 1, 2.
(v) End H
i
= Z Fin H
i
.
Note that H

1
is induced by if there is h H
2
such that = ( , h).
The set Fin H
i
of all endomorphisms of H
i
with nite rank image is an ideal of the
endomorphism ring EndH
i
and the last statement of the theorem means that this
ideal is a split extension in EndH
i
.
Hence each H
i
is separable and essentially indecomposable, this means any de-
composition H
i
= C E must have a summand E or C of nite rank. The key for
proving Theorem 1.2 are new algebraic and combinatorial methods and some old
techniques from earlier papers like [11] or [4].
2. Reflexive groups of cardinality 2
0
under Martins axiom
In contrast to the results in Section 3 concerning the existence of arbitrarily large
summands of reexive groups larger then a supercompact cardinality in this section
7
2
7


r
e
v
i
s
i
o
n
:
2
0
0
0
-
0
5
-
0
9







m
o
d
i
f
i
e
d
:
2
0
0
0
-
0
5
-
0
9


REFLEXIVE SUBGROUPS OF THE BAER-SPECKER GROUP AND MARTINS AXIOM 3
we will now construct essentially indecomposable reexive groups under Martins
axiom MA. As above let P =

n
e
n
Z be the Baer-Specker group of all elements
P = x =

i
x
i
e
i
: (x
i
Z).
Here e
i
can be viewed as the element x with coecients x
ij
=
ij
the Kronecker
symbol. Hence
S = e
i
: i ) =

i
e
i
Z
is a subgroup of P of all elements x of nite support
[x] = i : x
i
,= 0
and P/S is algebraically compact by an old result of Balcerzyk, see Fuchs [8].
Obviously P/S is torsion-free or equivalently S is pure in P. Pure subgroups
X P are denoted by X

P. Moreover, let D be the Z-adic closure of S in P,

hence D/S is the maximal divisible (torsion-free) subgroup of P/S which has size
2
0
. If H is an abelian group, then Fin H denotes the ideal of all endomorphisms
EndH with Im of nite rank. The groups we want to construct will be
sandwiched between S and D.
We will use Martins axiom for -centered sets, which is a (proper) consequence
of the well-known Martins axiom and equivalent to the combinatorial principle
P(2
0
) (below) as shown by Bell [2]. Recall that D P is dense in the poset P
if for any p P there exists d D such that p d. Martins axiom is based
on posets P with c.c.c. using that p, q P are compatible if there is r P with
p, q r. Recall that F P is bounded by r, say F r if f r for all r F. A
set X P is directed if all nite subsets of X are bounded in X and X is called
-centered (-directed) if it is the countable union of directed subsets. Replacing
c.c.c. by -centered MA turns into Martins axiom for -centered sets:
Let D be a collection of dense subsets D of the poset P. If [D[ < 2
0
and (P, )
is a -centered poset then there is a D-generic subset G P. Hence G is directed
and meets every D D : G D ,= .
See [7, p. 164] for MA with c.c.c. Note that the main result in Bell [2] is that
Martins axiom for -centered sets is equivalent to
The combinatorial principle P(2
0
): If D is a collection of subsets of such
that [D[ < 2
0
and

F is innite for every nite F D, then there is an innite
B such that B D is innite for all D D.
Martins axiom will help us to dene a scalar product or bilinear form on
suitable pairs H = (H
1
, H
2
) of pure subgroups H
j
of D. We begin with
: S S Z with (e
i
, e
j
) =
ij
.
Hence is the unique integer valued, bilinear form on SS. It extends by continuity
uniquely to the non-degenerated, symmetric bilinear form
: D D

Z with

Z the Z-adic completion of Z.
7
2
7


r
e
v
i
s
i
o
n
:
2
0
0
0
-
0
5
-
0
9







m
o
d
i
f
i
e
d
:
2
0
0
0
-
0
5
-
0
9


4 R

UDIGER G

OBEL AND SAHARON SHELAH

We keep this map xed though out this section and also denote restrictions to pairs
of subgroups by . Note that

Z is the cartesian product of the additive groups of
p-adic integers over all primes p and if
a =

i
a
i
e
i
D and b =

i
b
i
e
i
D, then (a, b) =

i
a
i
b
i
is well-dened and symmetry (a, b) = (b, a) is obvious. Now we consider such
pairs H = (H
1
, H
2
) such that (H
1
, H
2
) takes only values in Z. More precisely,
let H P if and only if the followings holds for j = 1, 2:
(i) S H
j

D
(ii) [H
j
[ < 2
0
(iii) : H
1
H
2
Z.
We dene a partial order on P and say
Denition 2.1. If H, H

P then H H

if and only if H
1
H

1
and H
2
H

2
.
The next crucial lemma of this paper will show under MA that P is a rich
structure.
Main Lemma 2.2. (ZFC + MA) Let H = (H
1
, H
2
) P, b P D and b
n
H
1
for n . Then there is a =

i
a
i
e
i
D such that for H

1
= H
1
, a)

D and
H

= (H

1
, H
2
) the following holds.
(i) H H

P and (a, b)

Z Z.
(ii) (a) Either

i
a
i
b
i
/ H

1
(b) or there is t Z such that b
j
te
j
: j ) is a free direct summand
of nite rank.
Remark. By symmetry we obtain a dual result of the Main Lemma 2.2 with
a H

2
and (b, a)

Z Z and (ii) accordingly.

Proof. Let b =

i
b
i
e
i
P D and H = (H
1
, H
2
) P be given by the
lemma. Moreover we assume that condition (ii)(b) of the lemma does not hold. This
is to say that we must show (ii)(a) of the lemma. This implication will follow at the
end of the proof from density of the sets D
5
dtn0
and density will be a consequence
of the assumption just made.
We want to approximate a H

1
by a forcing notion F, a partially ordered set,
used for application of MA. The elements p F are triples
(M
p
, A
p
, n
p
) with A
p
= a
p
l
: l < l
p
), M
p
= m
p
x
=

l<l
p
x
l
a
p
l
: x =

i
x
i
e
i
u
p

subject to the following conditions.

(i) u
p
is a nite subset of H
2
,
(ii) l
p
, a
p
l
, m
p
x
Z, and n
p
N.
We call l
p
the length of the nite sequence of integers A
p
and note that n[m means
n divides m in Z. In order to turn F into a partially ordered set let p q for some
p, q F if the following holds.
u
p
u
q
, l
p
l
q
, A
p
= A
q
l
p
,
n
p
[n
q
, and if l
p
l < l
q
then n
p
[a
q
l
,
if x =

l
x
l
e
l
u
p
then m
q
x
= m
p
x
=:

l<l
p
x
l
a
p
l
or equivalently

l
p
l<l
q
x
l
a
p
l
= 0.
7
2
7


r
e
v
i
s
i
o
n
:
2
0
0
0
-
0
5
-
0
9







m
o
d
i
f
i
e
d
:
2
0
0
0
-
0
5
-
0
9


REFLEXIVE SUBGROUPS OF THE BAER-SPECKER GROUP AND MARTINS AXIOM 5
If p, q F then let
p q (l
p
= l
q
, A
p
= A
q
, n
p
= n
q
)
and note that is an equivalence relation on F. If p F then let F
p
= q F :
q p. Surely F decomposes into countably many of such uncountable equivalence
classes F
p
. We claim that each of them is directed. If q
1
, q
2
F
p
then n
qi
=
n
p
, A
pi
= A
p
, l
pi
= l
p
, hence q
i
= (M
qi
, A
p
, n
p
), and if x =

i
x
i
e
i
u
q1
u
q2
,
then
m
q1
x
=

i<l
q
1
x
i
a
q1
i
=

i<l
p
x
i
a
p
i
= m
q2
x
.
If we dene q

F by u
q

= u
q1
u
q2
, A
q

= A
p
, l
q

= l
p
, n
q

= n
p
, then M
q

=
m
q

x
=

i<l
p x
i
a
p
i
: x u
q

= M
q1
M
q2
, hence q

= (M
q

, A
p
, n
p
) is a member
of F and q
1
, q
2
q

. The claim is shown and by denition

(F, ) is a -centered poset, (2.1)
as required for applications of MA for -centered sets. In order to apply MA
eectively we must dene dense subsets of F which describe local properties of the
desired a D. If x =

i
x
i
e
i
H
2
, m N, l
0
then let
D
1
x
= p F : x u
p
, D
2
m
= p F : m[n
p
, D
3
l0
= p F : l
0
l
p

,
D
4
m
= p F :

l<l
p
b
l
a
p
l
, m mod n
p

and for d D, t Z and n

0
N let
D
5
dtn0
= p F : m N (m[n
p
, n
0

i<l
p
a
p
i
b
i
t

i<l
p
a
p
i
e
i
d , 0 mod mD.
First note that we dened < 2
0
subsets of F as required for MA. Next we want
to show that all these sets are dense in F. The rst three cases are easy while
the remaining two cases need work. For D
1
x
with x =

i
x
i
e
i
we take any
p F and dene q like p just enlarging u
q
= u
p
x, let m
q
x
=

l<l
p
x
l
a
p
l
and
enlarge M
q
= m
p
y
: y =

i
y
i
e
i
u
p
m
q
x
as well, hence q q and D
1
x
is dense in F. Similarly take any p q F with m[n
q
, hence D
2
m
is dense. For
D
3
l0
replace any A
p
by A
q
= (A
p
)

(0, . . . , 0) with (0, . . . , 0) a vector of l

0
zeros
and let u
q
= u
p
, l
q
= l
p
+ l
0
, n
q
= n
p
. In the fourth case we rst notice that
b =

i
b
i
e
i
P D by hypothesis, hence there is s

N such that the set

W = k : b
k
Z s

Z is innite.
Suppose p F contradicts the density of D
4
m
for some m N, hence
there is no q D
4
m
with p q. (2.2)
We write
u
p
= a
1
, . . . a
k1
H
2
and let a
j
=

i
a
ji
e
i
.
7
2
7


r
e
v
i
s
i
o
n
:
2
0
0
0
-
0
5
-
0
9







m
o
d
i
f
i
e
d
:
2
0
0
0
-
0
5
-
0
9


6 R

UDIGER G

OBEL AND SAHARON SHELAH

Also consider the k -matrix (s ) (G)
_
_
_
_
_
_
_
a
11
a
12
. . . a
1s
. . .
a
21
a
22
. . . a
2s
. . .
.
.
.
a
k1,1
a
k1,2
. . . a
k1,s
. . .
b
1
b
2
b
s
. . .
_
_
_
_
_
_
_
as well as the (k 1) -matrix (H) deleting the last row of b
s
s of the matrix
above.
_
_
_
_
_
a
11
a
12
. . . a
1s
. . .
a
21
a
22
. . . a
2s
. . .
.
.
.
a
k1,1
a
k1,2
. . . a
k1,s
. . .
_
_
_
_
_
We pick nite subsets w of [l
p
, ) and consider the column vectors g
p
l
(l w) of the
rst matrix (G) and h
p
l
(l w) of the second matrix (H) accordingly and claim
that for all nite
w [l
p
, ) and d
l
Q [

lw
d
l
h
p
l
= 0

lw
d
l
g
p
l
= 0]. (2.3)
The proof is trivial. For suppose for contradiction that

lw
d
l
h
p
l
= 0 but

lw
d
l
g
p
l
,= 0
for some nite w [l
p
, ) and d
l
Q. Hence

lw
d
l
a
jl
= 0 for j < k and v =

lw
d
l
b
l
,= 0. (2.4)
Multiplying this homogeneous system of equations and the inequality by a large
enough natural number we may assume that
d
l
n
p
Z for all l w.
We now want to dene q > p with q D
4
m
and distinguish two cases. If

l<l
p
b
l
a
p
l
,=
m then choose n
q
large enough such that n
p
[n
q
and

l<l
p
b
l
a
p
m , 0 mod n
q
and put u
p
= u
q
, M
p
= M
q
, A
p
= A
q
. Then p < q and

l<l
q b
l
a
q
l
, m mod n
q
hence q D
4
m
is a contradiction, see (2.2). If

l<l
p b
l
a
p
l
= m, then choose l
q
>
sup(w l
p
) and dene q such that
a
q
l
(t) =
_
_
_
a
p
l
(t) if t [0, l
p
)
d
l
if l w
0 if l [l
p
, ) w.
Set u
q
= u
p
H
2
and using (2.4) let n
q
be large enough such that n
p
[n
q
but v , 0
mod n
q
. It follows p < q and

l<l
q
b
l
a
q
l
=

l<l
p
b
l
a
q
l
+

lw
b
l
d
l
= m+v.
Hence q D
4
m
is another contradiction, see (2.2). The linear dependence (2.3)
between the h
p
l
s and g
p
l
s is shown. Now we want to use (2.3) to derive a nal
contradiction for (2.2). For each nite w we have a Q-vector space V
w
=
h
p
l
: l w) of nite dimension k. Hence there is an r and a nite
w

[l
p
, ) such that h
p
l
(l w

) is a maximal independent set - and V

w
has
7
2
7


r
e
v
i
s
i
o
n
:
2
0
0
0
-
0
5
-
0
9







m
o
d
i
f
i
e
d
:
2
0
0
0
-
0
5
-
0
9


REFLEXIVE SUBGROUPS OF THE BAER-SPECKER GROUP AND MARTINS AXIOM 7
maximal dimension [w

[ = r k. If w

w [l
p
, ) for some nite w, then the
sub-matrix (H
w
) = (h
p
l
, l w) of (H) has nite column rank r, hence row rank r
as well and there is a subset z 1, . . . , k 1 of size r such that
a
j
w : j z is maximal independent.
By (2.3) b w is a linear combination of the a
j
w : j z and there are unique
elements c
l
Q, l z such that b w =

lz
c
l
a
l
w. If we increase w we have
the same coecients by maximal independence. Hence
b [l
p
, ) =

lz
c
l
a
l
[l
p
, ). (2.5)
We can choose m

N large enough such that m

c
l
s

Z for all l z. If t W is
large enough, then m

[a
lt
for all l z. Using (2.5) we get
b
t
=

lz
c
l
a
lt
s

Z
contradicting W. Hence D
4
m
is dense in F.
In order to show density of the last collection of subsets, suppose there are
d H
1
, t Z and n
0
N such that
D
5
dtn0
is not dense in F. (2.6)
Hence there is p F such that
no q D
5
dtn0
satises p q. (2.7)
Let u
p
= c
i
=

j
c
i
j
e
j
: i < k and l
p
< l < . We want to consider extensions
p q with l
q
= l and hence let
F
l
= (y
l
p, . . . , y
l1
) Z
llp
:
l1

j=l
p
c
i
j
y
j
= 0, i < k
which is a non-trivial subgroup of the free group Z
llp
for any large enough l. Also
let
s(y
l
p, . . . , y
l1
) = n
0
(

i<l
p
a
p
i
b
i
+
l1

i=l
p
y
i
b
i
) t(

i<l
p
a
p
i
e
i
+
l1

i=l
p
y
i
e
i
) d.
We claim that
(y
l
p, . . . , y
l1
) F
l
s(y
l
p , . . . , y
l1
) = 0 holds in D (2.8)
If s(y
l
p, . . . , y
l1
) ,= 0 for some (y
l
p , . . . , y
l1
) F
l
, then there is some m N
such that
s(y
l
p, . . . , y
l1
) , 0 mod mD. (2.9)
We now dene some q F taking
l
q
= l, n
q
= n
p
m, u
q
= u
p
, M
q
= m
q
x
=

i<l
q
x
i
a
q
i
: x u
q

where
a
q
i
=
_
a
p
i
if i < l
p
y
i
if l
p
i < l.
7
2
7


r
e
v
i
s
i
o
n
:
2
0
0
0
-
0
5
-
0
9







m
o
d
i
f
i
e
d
:
2
0
0
0
-
0
5
-
0
9


8 R

UDIGER G

OBEL AND SAHARON SHELAH

Clearly q F and also q D
dtn0
from (2.9), hence p , q from (2.7). On the
other hand

l1
j=l
p
c
i
j
a
q
j
= 0 from F
l
and denition of a
q
i
would imply p q, a
contradiction which proves the claim (2.8).
If we let
s
i
=

j
s
i
j
e
j
= n
0
b
i
te
i
D (l
p
i < ),
then the implication of (2.8) can be written as
l1

i=l
p
y
i
s
i
= d +t

i<l
p
a
i
e
i
n
0

i<l
p
a
p
i
b
i
.
From (0, . . . , 0) F
l
follows
n
0

i<l
p
a
p
i
b
i
= d +t

i<l
p
a
i
e
i
(2.10)
and from
(y
l
p, . . . , y
l1
) F
l
also follows
l1

i=l
p
y
i
s
i
= 0. (2.11)
If we view s
i
=

j
s
i
j
e
j
as innite row vector (l
p
i < l), then from the matrix
_
_
_
_
_
s
l
p
0
s
l
p
1
. . . s
l
p
k
. . .
s
l
p
+1
0
s
l
p
+1
1
. . . s
l
p
+1
k
. . .
.
.
.
s
l1
0
s
l1
1
. . . s
l1
k
. . .
_
_
_
_
_
we have nite column vectors s
n
= (s
i
n
: l
p
i < l) for any n . Let c
i
[l
p
, l) be
the restriction of c
i
viewed as an innite column vector restricted to the coordinates
j such that l
p
j < l, then
c
i
[l
p
, l) : i < k)
denotes the vector space over Q generated by these nite column vectors. We claim
that
s
n
c
i
[l
p
, l) : i < k) for all n .
Naturally F
l
Z
ll
p
Q
ll
p
. If F
l
= F
l
) denotes the subspace of Q
ll
p
generated
by F
l
, then F
l
= c
i
[l
p
, l) : i < k)

where orthogonality is dened naturally by

U

= x Q
ll
p
: x u = 0 u U
for U Q
ll
p
and the obvious scalar product x u =

ill
p
x
i
u
i
. From (2.11)
follows
F
l
= s
n
: n )

.
Using again, we have
s
n
: n )

c
i
[l
p
, l) : i < k)

which is
s
n
: n ) c
i
[l
p
, l) : i < k)
as dimQ
ll
p
is nite. This shows the claim.
7
2
7


r
e
v
i
s
i
o
n
:
2
0
0
0
-
0
5
-
0
9







m
o
d
i
f
i
e
d
:
2
0
0
0
-
0
5
-
0
9


REFLEXIVE SUBGROUPS OF THE BAER-SPECKER GROUP AND MARTINS AXIOM 9
Now let l be large enough such that c
i
[l
p
, l) : i < k) has maximal dimension
k

k and let c
i
[l
p
, l) (i < k

) be a basis of this vector space. We now can write

s
n
=

i<k

r
nl
i
c
i
[l
p
, l)
with unique coecients r
nl
i
Q. By uniqueness these coecients are independent
of l for any larger l, say that r
nl
i
= r
n
i
. In the system of equations
s
n
=

i<k

r
n
i
c
i
[l
p
, l), (l
p
l < , n )
we can also eliminate l and get
s
n
=

i<k

r
n
i
c
i
[l
p
, ), n .
From s
j
and b
j
=

n
b
j
n
e
n
we have that s
j
n
= n
0
b
j
n
t
jn
=

i<k
r
n
i
c
i
n
for any
n l
p
, hence (n
0
b
j
te
j
) [l
p
, ) c
i
[l
p
, ) : i < k

) and
U = n
0
b
j
te
j
: j )

D
has nite rank. Hence U is a free direct summand of D, see Fuchs [8]. If n
0
does
not divide t, then modulo n
0
D the image of U is te
j
+ n
0
D : j )

and has
innite rank, which is impossible. Hence n
0
[t and we rename tn
1
0
by t. Using
purity, we get that U = b
j
te
j
: j )

is a free direct summand of D which

contradicts our assumption that condition (ii)(b) does not hold. Hence D
dtn0
is
dense in F indeed, see (2.6).
We are ready to apply Martins axiom. There is a generic set G F which meets
the dense subsets of F just constructed. We dene a =

i
a
i
e
i
such that a
i
= a
p
i
for any p G with i < l
p
. Here we applied D
3
l0
and note that G is directed, hence
a is well-dened. Also a D by D
2
m
. Let H

1
= H
1
, a)

D be the pure subgroup

of D generated by H

1
= H
1
+Za and H

= (H

1
, H
2
). Then clearly H H

and we
claim that H

P. It is enough to show (iii) for P. If c H

1
then c = ka + e
for some k N, e H
1
. If y H
2
, then consider (c, y) = k(a, y) + (e, y).
From density of D
1
y
and p D
1
y
G and the choice of a follows (a, y) = m
p
y
Z
and therefore (c, y) Z. The map extends to H

1
H
2
Z. If x H

1
then tx = h H

1
for some t N and if x =

i
x
i
e
i
, h =

i
h
i
e
i
then
tx =

i
tx
i
e
i
=

i
h
i
e
i
and h
i
= tx
i
for all i . Hence
(h, y) = (tx, y) =

i
tx
i
y
i
= t(

i
x
i
y
i
) = t(x, y) t

Z Z
and by purity of Z

Z also t(x, y) tZ and by torsion-freeness (x, y) Z.
We have seen that H

P. Next we claim that

by denition of a and b we have z = (a, b) =

i
b
i
a
i

Z Z. (2.12)
Note that a
i
0 in the Z-adic topology, hence b
i
a
i
0 and z

Z is well-
dened. If z Z and n N then

i<k
b
i
a
i
z mod n for any large enough k,
which contradicts D
4
|z|
.
7
2
7


r
e
v
i
s
i
o
n
:
2
0
0
0
-
0
5
-
0
9







m
o
d
i
f
i
e
d
:
2
0
0
0
-
0
5
-
0
9


10 R

UDIGER G

OBEL AND SAHARON SHELAH

Finally we show that

i
a
i
b
i
/ H

1
. Otherwise there are t, n N and d H
1
such that
n

i
a
i
b
i
ta d = 0. (2.13)
Let p G D
5
dtn
from density of D
5
dtn
and choose m from the denition of D
5
dtn
.
Hence
n

i<l
p
a
p
i
b
i
t

i<l
p
a
p
i
e
i
d D mD.
On the other hand a
p
i
= a
i
for all i < l
p
from p G and m[n
p
by p D
5
dtn
.
The set G is directed, hence m[a
i
for all i l
p
. so n

il
p a
i
b
i
mD as well as
t

il
p
a
i
e
i
mD. The last displayed expression becomes n

i
a
i
b
i
ta d
D mD which contradicts (2.13). The Main Lemma 2.2 is shown.
From the proof of the Main Lemma 2.2 we have an immediate
Corollary 2.3. If H = (H
1
, H
2
) P, a D with (a, y) Z for all y H
2
and
H

1
= H
1
, a)

D then (H

1
, H
2
) P, in particular : H

1
H
2
Z.
In order to show Theorem 1.2 we want to use an ad hoc and preliminary de-
nition. Here we also use that is symmetric.
Denition 2.4. A pair H = (H
1
, H
2
) of pure subgroups of D is a full pair if the
following holds.
(i) There is an increasing continuous chain H

= (H
1
, H
2
) P with 2
0
and union (H
1
, H
2
).
(ii) If b P D and d 1, 2 there is a H
d
such that (a, b)

Z Z.
(iii) If b D, then there is d 1, 2 and b H
d
or for some a H
3d
follows
(a, b)

Z Z.
(iv) If d 1, 2 and b
n
H
d
, (n ), there is a =

i
a
i
e
i
H
d
such that
(a) either

a
i
b
i
D H
d
or
(b) or there is t Z such that b
j
te
j
: j ) is a free direct summand
of nite rank.
.
Lemma 2.5. (ZFC + MA) There is a full pair H = (H
1
, H
2
).
Proof. Enumerate P D = b

: 2
0
and D

= (b
n

)
n
: 2
0
with
2
0
repetitions such that any element appears 2
0
times. We want to construct
the P-chain inductively and let (H
01
, H
02
) = (S, S). By continuity we only have
to dene H
+1
. Alternatively we switch between 1 and 2, say we are in case H
1
and consider b

and (b
n

)
n
. By the Main Lemma 2.2 there is a

D such
that (H
(+1)1
, H
2
) P where H
(+1)1
= H
1
, a

D and (a

, b

)

Z Z.
Moreover (b
n

)
n
satises condition (ii) of the Main Lemma 2.2 for b
n
= b
n

. A
dual argument (case 2) provides H
(+1)2
. Hence H
+1
= (H
(+1)1
, H
(+1)2
) P.
This nishes the construction of H and Denition 2.4 is easily checked.
Lemma 2.6. If H

1
for a full pair H = (H
1
, H
2
), then there is b H
2
with
= ( , b)
7
2
7


r
e
v
i
s
i
o
n
:
2
0
0
0
-
0
5
-
0
9







m
o
d
i
f
i
e
d
:
2
0
0
0
-
0
5
-
0
9


REFLEXIVE SUBGROUPS OF THE BAER-SPECKER GROUP AND MARTINS AXIOM 11
Remark A similar result holds for H

2
.
Proof. Let b
j
= e
j
Z for all j , and set b =

j
b
j
e
j
P. If
a H
1
D, then write a =

j
a
j
e
j
and by continuity a = (

j
a
j
e
j
) =

j
a
j
(e
j
) =

j
a
j
b
j
= (a, b). Hence = ( , b). If b P D, then by
Denition 2.4 there is x H
1
with x = (b, x)

Z Z contradicting H

1
,
hence b D. Similarly by Denition 2.4 (iii) we have b H
2
and the lemma
follows.
The pair H = (H
1
, H
2
) in Lemma 2.6 satises conditions (i) and (iii) of the
Theorem 1.2. Reexivity follows easily as in [9] or [10] because the dual maps
are induced by scalar multiplication. As a subgroup of P, each H
i
is
1
-free, see
Fuchs [8]. Slenderness can easily be checked and is left to the reader, hence (ii) of
Theorem 1.2 follows. Condition (iv) was added for sake of completeness, it was the
main goal in [9] or [10] and can be derived here using the arguments from there.
The nal condition (v) will follow immediately from our next Lemma 2.7.
Lemma 2.7. If H = (H
1
, H
2
) is a full pair and End H
1
then there is s Z
such that s1 Fin H
1
where Fin H
1
is the ideal of End H
1
of all endomorphisms
of nite rank.
Proof. If e
j
= b
j
, j then using that H is a full pair, we nd a =

i
a
i
e
i
H
1
such that Denition 2.4(iv) holds. By by continuity also
b = a = (

i
a
i
e
i
) =

i
a
n
b
i
H
1
which shows that we are in case (b) of Denition 2.4(iv). The subgroup U =
b
j
te
j
: j ) is a free direct summand of nite rank of D. However the image
of S =

i
e
i
Z under t id is in U, hence S( t id ) has nite rank, and by
continuity the same holds for H
1
( t id ), this is to say that t1 Fin H
1
.
3. Large reflexive groups
Let be a xed supercompact cardinal. Then there is a -complete (ne) ultra-
lter U over such that the constant function
j : V M = Ult (V, U) (x j(x)) (j(x)

= x for all )
is an elementary embedding of the universe V into the ultra power M; for details
see Kanamori [14, pp. 471, 298 - 306, 37 - 56]. If is a cardinal, then
H() = x V : [TC(x)[ <
is the set of all sets in V hereditarily < where TC(x) denotes the transitive closure
of the set x.
Theorem 3.1. If is a supercompact cardinal and H is a dual group of cardinality
, then for any < there is a direct summand H

of H with [H

[ < .
Then the following is immediate.
Corollary 3.2. Every reexive group of cardinality , with supercompact, has
arbitrarily large summands < .
7
2
7


r
e
v
i
s
i
o
n
:
2
0
0
0
-
0
5
-
0
9







m
o
d
i
f
i
e
d
:
2
0
0
0
-
0
5
-
0
9


12 R

UDIGER G

OBEL AND SAHARON SHELAH

Proof of Theorem 3.1: Let H = G

= Hom(G, Z) be as in the theorem. If

[G[ =
1
, [H[ =
2
then let > 2
1+2
and assume G =
1
, H =
2
as sets and
< . If P = P

(H()) is the poset of all subsets of H() of cardinality < ,

then by the above there is a -complete (normal and ne) ultralter D on P with
elementary embedding
(H(), ) M := Ult (P, D).
From H = G

each h H gives rise to a homomorphism

(h, ) : G Z
and : H G Z is a bilinear form. Moreover
(h, ) = 0 h = 0,
hence is not degenerated. Let C be the set of all N P subject to the conditions
(i) G, H, N
(ii) + 1 N
(iii) N is an elementary submodel of (H(), ).
(iv) If = otp(N ) is the order type of N , then (N, ) is isomorphic to
(H(), )), say by an isomorphism j
N
.
By supercompactness C D and by Loss theorem ([14, p. 47, Theorem 5.2] the
desired properties of H() carry over to N. If N C then dene
H

= H N and G

= G N.
From + 1 N P and + 1
1
= G, + 1
2
= H follows + 1 H

and
+ 1 G

, hence
[H

[ < and [G

[ <
and by (iii)
H

H, G

G are subgroups. (3.1)

Similarly, if

= H

, then

: H

Z
and from (iii) and

we have
H

= G

.
We are ready to use an old trick from functional analysis to show that H

is also a
summand of H. Let
G

= h H : (h, G

) = 0 where (h, G

) = (h, g) : g G

.
Clearly G

H, and consider any h H

. We have (h, G

) = 0 and from
h H

follows that in the submodel N the following holds

N [= (x G

N
(h, x) = 0).
By (iii) we also have
(H(), ) [= (x G (h, x) = 0),
hence (h, ) = 0 and h = 0 because is not degenerated. We conclude
H

= 0, G

H.
In order to show
H

+G

= H (3.2)
7
2
7


r
e
v
i
s
i
o
n
:
2
0
0
0
-
0
5
-
0
9







m
o
d
i
f
i
e
d
:
2
0
0
0
-
0
5
-
0
9


REFLEXIVE SUBGROUPS OF THE BAER-SPECKER GROUP AND MARTINS AXIOM 13
we consider any h H = G

and let = (h, ) G

which belongs to G

. From
(3.1) we nd h

such that (h

, ) = . If g

we have
(h h

, g

) = (h, g

) (h

, g

) = g

= 0,
hence h h

and h H

+ G

, and (3.2) follows. All together we see that

H

is a summand of H of the right size.

References
[1] H. Bass, Finitistic dimension and a homological generalization of semi-primary rings, Trans-
act. Amer. Math. Soc. 95, 466 488 (1960).
[2] M. G. Bell, On the combinatorial principle P(c), Fundamenta math. 114 (1981) 149 157.
[3] K. Devlin, S. Shelah, A weak version of which follows from 2

0
< 2

1
, Israel J. Math. 6,
239 247 (1978).
[4] M. Dugas, J. Irvin, S. Khabbaz, Countable rings as endomorphism rings, Quart. J. Math.
Oxford (2) 39 (1988), 201211.
[5] K. Eda, On Z-kernel groups, Archiv der Mathematik 41, 289 293 (1983).
[6] K. Eda, H. Ohta, On abelian groups of integer-valued continuous functions, their Z-dual and
Z-reexivity, in Abelian Group Theory, pp. 241 257, Gordon and Breach, London 1986.
[7] P. Eklof, A. Mekler, Almost free modules, Set-theoretic methods, North-Holland, Amsterdam
1990.
[8] L. Fuchs, Innite abelian groups - Volume 1,2 Academic Press, New York 1970, 1973.
[9] R. Gobel, S. Shelah, Some nasty reexive groups, to appear in Mathematische Zeitschrift
[10] R. Gobel, S. Shelah, Decompositions of reexive modules, to appear in Archiv der Mathematik
[11] R. Gobel, B. Wald, Martins axiom implies the existence of certain growth types, Mathema-
tische Zeitschrift 172 (1980), 107 121.
[12] R. Gobel, B. Wald, Separable torsionfree modules of small type, Houston Journal of Math.
16 (1990), 271 287.
[13] T. Jech, Set theory, Academic Press, New York 1978
[14] A. Kanamori, The higher innite, Springer, Berlin 1994.
R udiger G obel
Fachbereich 6, Mathematik und Informatik
Universit at Essen, 45117 Essen, Germany
email: R.Goebel@Uni-Essen.De
and
Saharon Shelah
Department of Mathematics
Hebrew University, Jerusalem, Israel
and Rutgers University, Newbrunswick, NJ, U.S.A
e-mail: Shelah@math.huji.ae.il