6

2
1


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
8
-
0
3







m
o
d
i
f
i
e
d
:
1
9
9
9
-
0
8
-
0
3


A NON-REFLEXIVE WHITEHEAD GROUP
PAUL C. EKLOF AND SAHARON SHELAH
Abstract. We prove that it is consistent that there is a non-reexive Whitehead
group, in fact one whose dual group is free. We also prove that it is consistent
that there is a group A such that Ext(A, Z) is torsion and Hom(A, Z) = 0. As an
application we show the consistency of the existence of new co-Moore spaces.
0. Introduction
This paper is motivated by a theorem and a question due to Martin Huber. He
proved [8] in ZFC that if A is
1
-coseparable (that is, Ext(A, Z
()
) = 0), then A is
reexive (that is, the natural map of A to its double dual A

= Hom(Hom(A, Z), Z) is
an isomorphism). He asked whether it is provable in ZFC that every Whitehead group
A (i.e., Ext(A, Z) = 0) is reexive. This is true in any model where every Whitehead
group is free. It is also true for Whitehead groups of cardinality
1
in a model of MA
+ CH (because they are
1
-coseparable: cf. [4, Cor. XII.1.12]). Moreover, it is true in
the original models of GCH where there are non-free Whitehead groups (cf. [12], [13], [4,
Thm. XII.1.9]). It was left as an open question in [4, p. 455] whether every Whitehead
group is reexive. Here we give a strong negative answer:
Theorem 0.1. It is consistent with ZFC that there is a strongly non-reexive strongly

1
-free Whitehead group A of cardinality
1
.
A group A is strongly non-reexive if A is not isomorphic to A

. In fact, the example

A has the property that A

is free of rank
2
(i.e., isomorphic to Z
(2)
) so A

is
isomorphic to the product Z
2
; it is therefore not isomorphic to A since its cardinality
is 2
2
>
1
. (See Theorem 1.5 and Corollary 1.6 of section 1.)
If Ext(A, Z) = 0, then A is separable ([5, Thm 99.1]) and hence A

is non-zero.
However, using the same methods we can also prove:
Theorem 0.2. It is consistent with ZFC that there is a non-free strongly
1
-free group
A of cardinality
1
such that Ext(A, Z) is torsion and Hom(A, Z) = 0.
It is not a theorem of ZFC that there is a non-free torsion-free group A such that
Ext(A, Z) is torsion. Indeed, in any model where every Whitehead group is freea
hypothesis which is consistent with CH or CH (cf. [11])if A is not free, then Ext(A, Z)
is not torsion ([7], [3], [4, Thm. XII.2.4]).
Theorems 1.5 and 0.2 provide new examples of possible co-Moore spaces (see section
6). In particular, we answer a question in [6, p. 46] by showing that it is consistent that
1991 Mathematics Subject Classication. Primary 20K35, 20K20, 03E35, 03E75; Secondary 13L05,
18G15, 55N10.
Key words and phrases. Whitehead group, reexive group, dual group, co-Moore space.
Partially supported by NSF Grants DMS-9501415 and DMS-9704477.
Research of second author supported by German-Israeli Foundation for Scientic Research & Devel-
opment Grant No. G-294.081.06/93. Pub. No. 621.
1
6
2
1


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
8
-
0
3







m
o
d
i
f
i
e
d
:
1
9
9
9
-
0
8
-
0
3


2 PAUL C. EKLOF AND SAHARON SHELAH
for any n 2 there is a co-Moore space of type (F, n) where F is a free group of rank

2
.
The models for both theorems result from a nite support iteration of c.c.c. posets and
are models of ZFC + CH. (Other methods will be needed to obtain consistency with
CH.) We begin the iteration with a poset which yields generic data from which the
group A is dened; we then iterate the natural posets which insure that Ext(A, Z) = 0
(resp. Ext(A, Z) is torsion). The hard work is in proving that Hom(A, Z) is as claimed.
We dene the forcing and the group more precisely in the next section and then prove
their properties in the succeeding sections.
1. The basic construction
The group-theoretic construction is a generalization of that in [4, XII.3.4]. Let E be a
stationary and co-stationary subset of
1
consisting of limit ordinals, and for each E,
let

be a ladder on , that is, a strictly increasing function

: whose range
approaches . Let F be the free abelian group with basis x

:
1
z
,n
:
E, n . Let g be a function from E to the integers 1. Let u be a function from
E to the subgroup x

:
1
) generated by x

:
1
such that u(, n) belongs
to x

: <

(n)). Let K be the subgroup of F generated by w

,n
: E, n
where
(1.1) w
,n
= 2
g(,n)
z
,n+1
z
,n
x

(n)
u(, n).
Let A = F/K. Then clearly A is an abelian group of cardinality
1
. Notice that because
the right-hand side of (1.1) is 0 in A, we have for each E and n the following
relations in A:
(1.2) 2
g(,n)
z
,n+1
= z
,n
+x

(n)
+u(, n)
and
(1.3) 2

n
j=0
g(,j)
z
,n+1
= z
,0
+
n

k=0
2

k1
j=0
g(,j)
(x

(k)
+u(, k))
Here, and occasionally in what follows, we abuse notation and write, for example, z
,n+1
instead of z
,n+1
+ K for an element of A. For each <
1
, let A

be the subgroup of
A generated by
(1.4) x

: < z
,n
: E , n .
Then, by (1.3), for each E, z
,0
+ A

is non-zero and divisible in A

+1
/A

by 2
m
for all m . Thus A
+1
/A

is not free and hence A is not free. (In fact (A)

E.)
Moreover, A is strongly
1
-free; in fact, for every <
1
, using Pontryagins Criterion
[4, IV.2.3] we can show that A/A
+1
is
1
-free for all
1
1.
We begin with a model V of ZFC where GCH holds, choose E V , and dene
the group A in a generic extension V
Q0
using generic ladders

, and generic u and g.

Specically:
Denition 1.1. Let Q
0
be the set of all nite functions q such that dom(q) is a nite
subset of E and for all dom(q), q() is a triple (
q

, u
q

, g
q

) where for some r

q

:

q

is a strictly increasing function: r

q

;
u
q

: r
q

:
1
) such that for all n < r
q

, u
q

(, n)

: <

(n)

;
and
6
2
1


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
8
-
0
3







m
o
d
i
f
i
e
d
:
1
9
9
9
-
0
8
-
0
3


A NON-REFLEXIVE WHITEHEAD GROUP 3
g
q

: r
q

n : n 1.
The partial ordering is dened by: q
1
q
2
if and only if q
1
q
2
; note that we follow
the convention that stronger conditions are larger. Clearly Q
0
is c.c.c. and hence E
remains stationary and co-stationary in a generic extension. We now do an iterated
forcing to make A a Whitehead group. We begin by dening the basic forcing that we
will iterate.
Denition 1.2. Given a homomorphism : K Z, let Q

be the poset of all nite

functions q into Z satisfying:
There are
0
<
1
< ... <
m
in E and r

: m such that dom(q) =

z

,n
: m, n r

: I
q

where I
q

1
is nite and is such that for all m
(1.5) n < r

u(

, n) x

: I
q
) and

(n) I
q
n < r

and for all m and n < r

, u(

, n) x

: I
q
) and
(1.6) (w

,n
) = 2
g(,n)
q(z

,n+1
) q(z

,n
) q(x

(n)
) q(u(

, n)).
(Compare with (1.1). The denition of q(u(

, n)) is the obvious one, given that q should

extend to a homomorphism.) Moreover, we require of q that for all ,= j in 0, ..., m,
(1.7)

j
(k) ,=

(i) for all k r

j
and i .
We will denote
0
, ...,
m
by cont(q) and r

by num(q,

). The partial ordering on

Q

is inclusion.
Proposition 1.3. (i) For every E and k , D
,k
= q Q

: cont(q) and
k num(q, ) is dense in Q

(ii) Q

is c.c.c.
Before proving Proposition 1.3, we prove a lemma:
Lemma 1.4. Given
0
, ...,
m
E, integers r

for m and a nite subset I

of
1
,
there are integers r

for all m and a nite subset I

of
1
containing I

such
that for all m:
(a)

(n) I

n < r

; and
(b) for all n < r

, u(

, n) x

: I

).
Proof. The proof is by induction on m 0. If m = 0 we can take
r

0
= maxr

0
, maxk + 1 :
0
(k) I

and take I

to be a minimal extension of I

0
(n) : n < r

0
satisfying (b); then (a)
holds because u(, n) x

: <

(n)). If m > 0, without loss of generality we can

assume that
0
<
1
< ... <
m
. Let
r

m
= maxr

m
, maxk + 1 :
m
(k) I

, mink :
m
(k) >
m1
.
As in the case m = 0, there exists

I containing I

such that (a) and (b) hold for

I for
= m. Then apply the inductive hypothesis to
0
, ...,
m1
,

I, and the r

( < m) to
obtain r

for < m and a minimal I

.
6
2
1


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
8
-
0
3







m
o
d
i
f
i
e
d
:
1
9
9
9
-
0
8
-
0
3


4 PAUL C. EKLOF AND SAHARON SHELAH
For q Q

and
1
, let q denote the restriction of q to
dom(q) (z
,n
: < , n x

: < ).
Say that occurs in q if cont(q) or x

dom(q).
proof of proposition 1.3. (i) Given E, k and p Q

, we need qp
such that q D
,k
. Let cont(p) =
0
, ...,
m
. We consider two cases. The rst is that
cont(p), that is, =
j
for some j m. We can assume that k > num(p,
j
). Apply
Lemma 1.4 with I

= I
p
, r

j
= k, and r

= num(p,

) for ,= j to obtain I

and r

. Then
we can dene q to be the extension of p with cont(q) = cont(p) and num(q,

) = r

and
I
q
= I

. Since (1.5) and (1.7) hold, we can inductively dene q(x

(i)
) and q(z

,i+1
)
for r

i < r

(setting q(x

) = 0 for I
q
rge(

: m if not already dened)

so that (1.6) holds. Note that (1.7) continues to hold.
The second case is when / cont(p). Let
m+1
= . Choose r

for m + 1 so
that r

num(p,

) for m and such that (1.7) holds, that is,

j
(n) ,=

(i) for
all n r

j
and i for all j ,= 0, ..., m + 1. Apply Lemma 1.4 to
0
, ...,
m+1
,
I
p
, and the r

to obtain r

for m + 1 and I

. Let I
q
= I

and num(q,

) = r

.
For m dene q(x

(i)
), q(z

,i+1
) and u(

, i) for num(p,

) i < r

by induction
on i as in the rst case. Dene q(z
,r

m+1
+1
) = 0 and dene q(z
,n
) for n r

m+1
by
downward induction, i.e.
q(z
,n
) = 2
g(,n)
q(z
,n+1
) q(x

(n)
) q(u(, n)) (w
,n
).
(Setting q(x

) = 0 where not already dened, we can assume q(x

(n)
) and q(u(, n))
are dened.)
(ii) Consider an uncountable subset q

:
1
of Q

. By the -system lemma we

can assume that cont(q

) :
1
forms a -system, i.e., there is a nite subset
of E such that for all ,= , cont(q

) cont(q

) = . By renumbering an uncountable
subset, we can assume that for all , if cont(q

), then > . Furthermore, by

passing to a subset and using (i) we can assume that if cont(q

) and

(n) < ,
then n < num(q

, ). By Fodors Lemma we can assume that there exists max

such that for all and n, if cont(q

) and

(n) < , then

(n) < and moreover

such that if I
q
and < , then < . We can also assume that for all , ,
q

= q

. If we pick < such that < and whenever occurs in q

,
then < , then we will have that q

. Notice that (1.7) will be satised: if

cont(q

) and cont(q

) and k num(q

, ) and m num(q

, ), then

(k) <

(m); moreover, if i and

(i) < , then

(i) < <

(k).
Similarly it follows that (1.5) holds.
Now P =

P
i
,

Q
i
: 0 i <
2

is dened to be a nite support iteration of length

2
so that for every i 1
Pi

Q
i
= Q

i
where
Pi

i
is a homomorphism: K Z and the
enumeration of names

i
: 1 i <
2
is chosen so that if G is P-generic and V [G]
is a homomorphism: K Z, then for some i 1,

i
is a name for in V
Pi
. Then P is
c.c.c. and in V [G] every homomorphism from K to Z extends to one from F to Z. This
means that Ext(A, Z) = 0, that is, A is a Whitehead group (see, for example, [4, p. 8]).
We claim moreover that:
Theorem 1.5. In V [G] A

(= Hom(A, Z)) is free of cardinality

2
.
6
2
1


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
8
-
0
3







m
o
d
i
f
i
e
d
:
1
9
9
9
-
0
8
-
0
3


A NON-REFLEXIVE WHITEHEAD GROUP 5
As a consequence we can conclude:
Corollary 1.6. In V [G] A is strongly non-reexive.
Proof. Since A

is isomorphic to Z
(2)
, A

is isomorphic to Z
2
and hence not isomorphic
to A because its cardinality is dierent. We remark also that A

is not slender, but A

is slender since it is a Whitehead group see [4, Prop. XII.1.3, p. 345]).
The next three sections are devoted to a proof of Theorem 1.5. The fact that A

has cardinality 2
1
is a consequence of a result of Chase [1, Thm. 5.6]; by standard
arguments it can be seen that in V [G] 2
0
= 2
1
=
2
. Let G

= p : p G, so
that G

is P

-generic. To prove that Hom(A, Z)

V [G]
is free, it suces to prove that:
(I) Hom(A, Z)
V [G1]
= 0;
(II) for every limit
2
, Hom(A, Z)
V [G

]
=

i<
Hom(A, Z)
V [Gi]
;
and
(III) for all i <
2
, Hom(A, Z)
V [Gi+1]
/ Hom(A, Z)
V [Gi]
is free, and in fact is either 0
or Z.
We shall prove (I) immediately, and then prove the other two parts in the next three
sections.
proof of (I): Notice rst that, by (1.3), if h Hom(A, Z) and h(x

) = 0 for all

1
, then h is identically zero. So suppose, to obtain a contradiction, that there
exists a Q
0
-name

h and p G
1
such that
p

h Hom(A, Z)

h(x

) = m
for some
1
and some non-zero integer m. Choose d such that 2
d
does not divide
m. For each E there exists p

p and c

Z such that
p

h(z
,0
) = c

.
By Fodors Lemma and a -system argument, there exist
1
,=
2
> such that c

1
=
c

2
, and if (for convenience of notation) we let p
i
= p

i
, r
p
1
1
= r
p
2
2
= r,
1
(n) =
2
(n),
u(
1
, n) = u(
2
, n) for all n < r and p
1
and p
2
are compatible. Then there is a condition
q Q
0
such that p
i
q for i = 1, 2 and
q
1
(r) =
2
(r) g(
1
, r) = d = g(
2
, r) u(
1
, r) = x

u(
2
, r) = 0.
Now consider a generic extension V [G

1
] where q G

1
. By subtracting (1.3) for n = r
and =
2
from (1.3) for n = r and =
1
and applying h we obtain that (in V [G

1
]) 2
d
divides h(u(
1
, r)) h(u(
2
, r)) = h(x

) h(0) = m. But this is a contradiction of the

choice of d.
2. Preliminaries
Before beginning the proof proper of (II) and (III), we prove a crucial Proposition
that we will need. For a xed m and S E, let Z
m
[S] denote the pure closure
in A of the subgroup generated by z
,m
+ K : S. For t , let Z
m,t
[S] denote
Z
m
[S] + 2
t
A.
Proposition 2.1. In V [G], for all m, t and all stationary S E, A/Z
m,t
[S] is a
nite group.
6
2
1


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
8
-
0
3







m
o
d
i
f
i
e
d
:
1
9
9
9
-
0
8
-
0
3


6 PAUL C. EKLOF AND SAHARON SHELAH
Proof. The proof is by contradiction. Suppose that q

G such that for some m, t

and some

S
q

|
P

S is a stationary subset of E and A/Z

m,t
[

S] is innite.
Let S

be the set of all E such that q

does not force /

S; then S

V is a
stationary subset of E. For each S

choose p

such that p

|

S.
We can assume that each p

satises:
() 0 dom(p

); dom(p

(0)); for each j dom(p

), p

(j) is a
function in V and not just a name; r
p

(0)

(= r

) is independent of
dom(p

(0)); if j dom(p

) 0, cont(p

(j)) implies dom(p

(0))
and num(p

(j), ) (= r

,j
) is r

and independent of . Moreover, if

dom(p

(0)) and > , then

(r

,j
) > .
When we say that occurs in p we mean that p() is non-empty, or occurs
in p(j) for some j > 0 or dom(p(0))

(n) : dom(p(0)), n < r

p(0)

, or
u(, n) / x

: ,= ) for some n < r

p(0)

. Without loss of generality we can assume

(passing to a subset of S

) that
() there exists such that for all S

, > and every ordinal <

which occurs in p

is less than ; dom(p

) : S

forms a -system,
whose root we denote C (i.e., dom(p

1
) dom(p

2
) = C = 0,
1
, ...,
d

for all
1
,=
2
in S

); r

(= r

) and r

,j
(= r

j
) are independent of .
Moreover, for every j C, dom(p

(j)) : S

forms a -system
and for all
1
,=
2
in S

, p

1
(j) and p

2
(j) agree on dom(p

1
(j))
dom(p
2
(j)), so p
1
(j)
1
= p
2
(j)
2
. Also, dom(p

(0)) and
p

(0) (dom(p

(0)) ) are independent of . Finally,

p

(0)

(n) (=
n
),
g
p

(0)
(, n) (= g(n)) and u
p

(0)
(, n) are independent of for each n < r

.
Let p

denote the heart of p

: S

; that is, dom(p

) = C and for all C,

dom(p

()) = dom(p

1
()) dom(p

2
()) (= C

, say) for
1
,=
2
S

; and p

() =
p

1
() C

= p

1
()
1
.
The conditions in () insure that if
1
<
2
are members of S

such that every ordinal

which occurs in p

1
is <
2
, then p

1
and p

2
are almost compatible; however, there may
be problems in determining a value for p

(j)(x

n
) for r

j
n < r

(independent of
= 1, 2); it is because of these that the following argument is necessary.
We can assume that r

m and that for all S

/ dom(p

(0)). Choose M t
such that g(n) M for all n < r

. Let
N = 2
(r

+1)M
.
To obtain a contradiction, it suces to prove that p

forces:
() A/Z
m,t
[S] is a group of cardinality N
d
This is a contradiction since p

. If p

does not force (), then there is a nite

subset of
1
and a condition p

such that p

forces
()(x

: ) +Z
m,t
[S])/Z
m,t
[S] has cardinality > N
d
.
(Note that it follows from (1.3) that A/p
t
A is generated by x

: ).) We can assume

that if occurs in p

, then . Let T be the subset of x

: ) composed of
all elements of the form

+ x

where 0 c

< 2
t
. Let = 2
||t
; so T has
> N
d
elements; list them as

: < . Now choose elements

: < of S

listed
in increasing order and such that the smallest,
0
, is larger than max and such that
6
2
1


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
8
-
0
3







m
o
d
i
f
i
e
d
:
1
9
9
9
-
0
8
-
0
3


A NON-REFLEXIVE WHITEHEAD GROUP 7
every ordinal which occurs in p

is <
+1
. Moreover, we can choose them so that for
any < , the common part of p

and p

is p

; that is, dom(p

) dom(p

) = C =
dom(p

) and for all C, dom(p

()) dom(p

()) = dom(p

()). (So, in particular,

/ dom(p

(0)).)
Choose new ordinals

for 1 < such that

1
<
0
<
0
<
1
<
1
< ... <

<
+1
<
+1
< ...
Moreover, we make the choice so that for all ,

is larger than any ordinal <

which
occurs in any p

k
. There is a condition q
0
Q
0
which extends p

(0) and each p

(0)
( < ) such that q
0
forces for all < :

(r

) =
1
;

(r

+ 1) =

; u(

, r

) =

; and g(

, r

) = t.
This q
0
will force versions of (1.5) and (1.7) for the

. Also choose q
0
to force values for

(r

) for any

<
(dom(p

(0))

) so that (1.5) and (1.7) hold for p

()
p

() for any ,

< and any C. We claim that

(IV.1) There is a subset W of 0, ..., 1 of size at least N
d
and
a condition q P

2
which extends p

and p

for every W and

satises q(0) = q
0
Assuming (IV.1), let us deduce a contradiction, which will prove that p

forces (_). Work

in a generic extension V [G

] such that q G

. For
1
,=
2
in W we have
1

2

x

: ) Z
m,t
[S] because
z

1
,r

2
,r
=
1

2
+ 2
t
a
for some a A and, letting e =

r

1
n=m
g(n),
2
e
(z

1
,r

2
,r
) = z

1
,m
z

2
,m
Z
m
[S]
so since Z
m
[S] is pure-closed,
1

2
+ 2
t
a Z
m
[S], and hence
1

2
Z
m,t
[S].
Therefore (x

: ) +Z
m,t
[S])/Z
m,t
[S] has cardinality at most
2
||t
/[W[ = /[W[ N
d
which is a contradiction of the choice of p

.
In order to prove (IV.1) we dene inductively, for 1 n d+1, a condition q
n
P

n
(where
n
is as in the enumeration of C for n d and
d+1
=
2
) such that q
n
p

n
and for n

< n, q
n

n
q
n
. We also dene a subset W
n
of W of size at least N
(n1)
such that for each W
n
, q
n
p

n
. (So in the end we let q = q
d+1
and W = W
d+1
.)
To begin, let W
1
= 0, ..., 1 and let q
1
be any common extension of q
0
and the
p

1
. (There is no problem nding such an extension.) Suppose now that q
n
and
W
n
have been dened for some n 1. Choose q
n
q
n
in P

n
such that q
n
decides
for all W
n
the value of

n
(w
,k
) for all dom(p

(0)
) and all k r

. For each
W
n
x s

V such that s

n
extends p

(
n
) and satises num(s

) = r

+1,
s

(x
1
) < 2
M
, and s

(x

k
) < 2
M
for all k < r

. (This is possible by the proof

of Proposition 1.3 since we only need to nd solutions to the equations (1.2) modulo
2
M
since g
p

(0)
(

, k) M for k r

.)
Dene an equivalence relation
n
on W
n
by:
1

n

2
i s
1
s
2
is a function. By
choice of M and N, there is an equivalence class, W
n+1
, of size at least [W
n
[/N. For
W
n+1
we can dene a common extension q
n+1
(
n
) of p

(
n
) and the p

(
n
) and
let q
n+1

n
= q
n
. This completes the inductive construction.
6
2
1


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
8
-
0
3







m
o
d
i
f
i
e
d
:
1
9
9
9
-
0
8
-
0
3


8 PAUL C. EKLOF AND SAHARON SHELAH
For any abelian group H, let (H) be the Chase radical of H: the intersection of
the kernels of all homomorphisms of H into an
1
-free group (cf. [2]). Then H/(H)
is
1
-free ([2, Prop. 1.2], [4, p. 290]). Let cl(Z
m
[S]) be dened by: cl(Z
m
[S])/Z
m
[S] =
(A/Z
m
[S]), so in particular A/ cl(Z
m
[S]) is
1
-free. Notice also that every homomor-
phism from A to Z is determined on cl(Z
m
[S]) by its values on z
,m
+K : S.
Corollary 2.2. In V [G], for all m and stationary S E, A/ cl(Z
m
[S]) is a nite
rank free group.
Proof. If not, then since A/ cl(Z
m
[S]) is
1
-free, it contains a free pure subgroup of
countably innite rank. Let a
n
+ cl(Z
m
[S]) : n be a basis of such a subgroup.
For any n ,= m, a
n
+ Z
m,1
[S] ,= a
m
+ Z
m,1
[S] since 2 does not divide a
n
a
m
mod
Z
m
[S] (or even mod cl(Z
m
[S])). Therefore a
n
+ Z
m,1
[S] : n is an innite subset
of A/Z
m,1
[S], which contradicts Proposition 2.1.
For the next Corollary we will need the following:
Lemma 2.3. The Chase radical, (H), of a torsion-free group H is absolute for generic
extensions.
Proof. We give an absolute construction of (H) using the fact that a torsion-free group is

1
-free if and only if every nite rank subgroup is nitely-generated (cf. [5, Thm. 19.1]),
that is, if and only if the pure closure of every nitely-generated subgroup is nitely-
generated. For any group H

, let (H

) be the sum of all nite rank subgroups G of H

which are not free but are such that every subgroup of G of smaller rank is free; it is easy
to see that for such G, (G) = G and hence (H

) (H

). Moreover, the denition

of (H

) is absolute. Now dene

(H) by induction:
0
(H) = 0,
+1
(H)/

(H) =
(H/

(H)),and for limit ordinals ,

(H) =
<

(H). It follows by induction that

(H) (H) for all

1
. We claim that (H) =

1
(H); it suces to prove that
H/

1
(H) is
1
-free. If not, then there is a nite rank subgroup of H/

1
(H) which is
not nitely-generated. We can choose one, G, of minimal rank, so all of its subgroups of
smaller rank are free; say G is the pure closure of a
1
+

1
(H), ..., a
n
+

1
(H); but then
for some <
1
, the pure closure of a
1
+

(H), ..., a
n
+

(H) is not free, but still has

the property that every subgroup of smaller rank is free; hence a
1
, ..., a
n

+1
(H),
which is a contradiction.
Corollary 2.4. If h Hom(A, Z)
V [G]
and for some i
2
, m and some stationary
set S V [G
i
], the sequence (h(z
,m
+ K) : S) belongs to V [G
i
], then h belongs to
V [G
i
].
Proof. Suppose h, S and m are as in the hypotheses. First we claim that h Z
m
[S]
belongs to V [G
i
]. Indeed we can dene h Z
m
[S] in V [G
i
] as follows: h(a) = k if
for some n ,= 0, na belongs to the subgroup generated by z
,m
+ K : S and
h(na) = nk; and otherwise h(a) = for some xed / Z. In fact, the second case never
occurs because h(a) is dened in V [G]. Next we claim that h cl(Z
m
[S]) belongs to
V [G
i
]. The proof is similar in principle, using the inductive construction of cl(Z
m
[S])
given by the proof of Lemma 2.3. But then by Corollary 2.2, h is determined by only
nitely many more values, so also h belongs to V [G
i
].
For the next corollary we introduce some notation that will be used in section 4. Let

i
V [G
i+1
] denote the generic function for Q
i
; thus
i
is a homomorphism: F Z
6
2
1


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
8
-
0
3







m
o
d
i
f
i
e
d
:
1
9
9
9
-
0
8
-
0
3


A NON-REFLEXIVE WHITEHEAD GROUP 9
extending
i
: K Z, where
i
is the interpretation in V [G
i
] of the name

i
. The
canonical map Hom(K, Z) Ext(A, Z) sends
i
to a short exact sequence
c
i
: 0 Z

B
i

A 0
and there is a commuting diagram
0 K

F A 0

1A
0 Z

B
i

A 0
where and

are inclusion maps. Moreover, for all z F, (

i
)(z) = z + K A.
Then
i
gives rise to a splitting
i
Hom(B
i
, Z)
V [Gi+1]
dened by
i
(
i
(z)) =
i
(z).
Thus
i
= 1
Z
(the identity on Z) and also
i

i
=
i
.
Corollary 2.5. If f Hom(B
i
, Z)
V [G]
and for some m and some stationary set
S V [G
i
], the sequence (f(
i
(z
,m
)) : S) belongs to V [G
i
], then f belongs to V [G
i
].
Proof. If Z

is dened to be the pure subgroup of B

i
generated by
i
(z
,m
) :
S (1) and C

is such that (B
i
/Z

) = C

/Z

, then induces an isomorphism of

C

/ rge() with cl(Z

m
[S]). Hence B
i
/C

= A/ cl(Z
m
[S]) is nite rank free; therefore,
arguing as in Corollary 2.4, f belongs to V [G
i
].
3. Proof of (II)
We divide the proof of (II) into three cases according to the conality of . The case
of conality
2
(i.e., =
2
) is trivial since any function from A (which has cardinality

1
) to Z must belong to V [G
i
] for some i < .
Let

h be a P

-name and p G

such that
p
P

h Hom(A, Z).
Then for each E there is p

and k

Z such that p

p and p

h(z
,0
+K) = k

.
Suppose that the conality of is , and x an increasing sequence (
n
: n )
whose sup is . Then there is n and a stationary subset S
1
of E, belonging to
V [G

], such that for S

1
, p

n
. Without loss of generality there is p

G
n
such
that p

forces
S = E : p

n
and k

Z s.t. p

h(z
,0
+K) = k

is stationary.
Then (h(z
,0
+ K) : S) belongs to V [G

n
], so h Hom(A, Z)
V [G

n
]
, by Corollary
2.4.
The nal, and hardest, case is when the conality of is
1
. Fix an increasing
continuous sequence (

: <
1
) whose sup is . Then there is
1
and a stationary
subset S of
1
such that for S, p

.
For any t 1, (Z
0,t
[S ] : <
1
) is a non-increasing sequence of groups. Since the
groups A/Z
0,t
[S] are nite, it follows that there is a countable ordinal
t
such that
for ,
t
, Z
0,t
[S ] = Z
0,t
[S ]. Therefore there is a countable ordinal

and a
countable subset Y of A such that for all t 1,
t

, and Y contains representatives

of all the elements of A/Z
0,t
[S

]. Increasing if necessary, we can assume that we

can compute h(y) in V [G

] for all y Y .
6
2
1


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
8
-
0
3







m
o
d
i
f
i
e
d
:
1
9
9
9
-
0
8
-
0
3


10 PAUL C. EKLOF AND SAHARON SHELAH
We claim that h belongs to V [G

]. In pursuit of a contradiction, suppose that there

are a A, conditions q
1
, q
2
P

/G

and integers c
1
,= c
2
such that q

h(a) = c

for = 1, 2. Choose t suciently large such that 2

t
does not divide c
2
c
1
and choose
<
1
such that q
1
, q
2
P

. For some y Y , ay Z
0,t
[S

] = Z
0,t
[S

]. Thus
a y = z + 2
t
a

for some a

A and z in the pure closure of the subgroup generated by z

j
,0
: j =
1, ..., n for some
1
, ...,
n
S

. For = 1, 2 there is an upper-bound r

of q

, p

1
, ..., p

n
. Then r
1
and r
2
force the same value, b, to h(z) (because they are
both p
j
for j = 1, ..., n) and the same value, k, to h(y) (because it is determined in
V [G

]). Therefore
r

2
t
divides c

k b.
So for = 1, 2, the integer c

k b is divisible by 2
t
. But this contradicts the choice
of t.
4. Proof of (III)
We continue with the notation from the end of section 2; so c
i
V [G
i
]. Suppose
that c
i
represents a torsion element of Ext(A, Z), of order e 1, that is, there is a
homomorphism g
i
: B
i
Z such that g
i
Z = e1
Z
, or more precisely, g
i
= e1
Z
. (We
consider the zero element to be torsion of order 1.) Then e
i
g
i
is a homomorphism from
B
i
to Z which is identically 0 on Z, so it induces a homomorphism
i
Hom(A, Z)
V [Gi+1]
(that is,
i
= e
i
g
i
) which is a new element of A

that is, it is not in V [G

i
].
To prove (III) it will suce to prove that if there is an element h of A

which is in
V [G
i+1
] but not in V [G
i
], then c
i
is torsion, and in that case h is an integral multiple
of
i
modulo (A

)
V [Gi]
.
Given such an h, let h

= h : B
i
Z. Clearly h

V [G
i+1
] V [G
i
]. We claim
that:
(III.1) For some integer c, h

c
i
belongs to V [G
i
].
Let us see rst why this Claim implies the desired conclusion. Note that c ,= 0 since
h

does not belong to V [G

i
]. Since (h

c
i
) Z = c1
Z
, we conclude that in V [G
i
], c
i
is torsion, of order e dividing c; let g
i
Hom(B
i
, Z)
V [Gi]
such that g
i
Z = e1
Z
. Let

i
be induced by e
i
g
i
, as above. Say c = ne; then h

c
i
+ng
i
belongs to V [G
i
] and
is identically 0 on Z so it induces a homomorphism f Hom(A, Z)
V [Gi]
. By composing
both sides with one sees that h = n
i
+f.
We shall now work on the proof of (III.1). Let F

be the subgroup of F generated by

x

: <
1
. We work in V [G
i
]. For any countable ordinal
1
E, dene
Q
i,
= q Q
i
: z
,n
dom(q) < and x

dom(q) < .
Then Q
i,
is a complete subforcing of Q
i
. In particular,
V [G
i+1
] = V [G
i
][G
i+1,
][H
i+1,
]
where G
i+1,
is Q
i,
-generic over V [G
i
] and H
i+1,
is Q
i
/G
i+1,
-generic over V [G
i
][G
i+1,
].
We claim:
(III.2) There is a countable ordinal
1
E such that in V [G
i
][G
i+1,
]
there is an assignment to every y F

of a function
y
: Z Z such
that for all y F

|
Qi/Gi+1,

h(y +K) =
y
(
i
(y)).
6
2
1


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
8
-
0
3







m
o
d
i
f
i
e
d
:
1
9
9
9
-
0
8
-
0
3


A NON-REFLEXIVE WHITEHEAD GROUP 11
Let us see rst why this implies (III.1). First we assert that the following consequence
of (III.2) holds in V [G
i
][G
i+1,
]:
(III.2.1) There is an integer c such that for every k Z, and every ,

(k)
x

(0) = kc.
To see this, let , with ,= , and let k Z. By the proof of Proposition 1.3,
there are conditions q
1
, q
2
Q
i
/G
i+1,
such that
q
1
|
i
(x

) = k
i
(x

) = 0
and
q
2
|
i
(x

) = 0
i
(x

) = k.
Let y = x

+x

. By (III.2) and the fact that

h and
i
are homomorphisms,
q
1
|
x

(k) +
x
(0) =
y
(k)
which implies that
x

(k) +
x
(0) =
y
(k) holds in V [G
i
][G
i+1,
]. Similarly, reasoning
with q
2
, we can conclude that
x
(k) +
x

(0) =
y
(k) holds in V [G
i
][G
i+1,
]. Thus

(k)
x

(0) =
x
(k)
x
(0) in V [G
i
][G
i+1,
]; we denote this value by (k). If we
can prove that for all k, (k) = k(1), then we can let c = (1). Again, let , with
,= and this time let y = kx

+ x

. Using conditions q
3
|
i
(x

) = 1
i
(x

) = 0
and q
2
|
i
(x

) = 0
i
(x

) = k we conclude that
k
x

(1) +
x
(0) = k
x

(0) +
x
(k)
from which it follows that k(1) = (k). This proves (III.2.1)
Now work in V [G
i+1
]; we have
h(x

+K) =
x

(
i
(x

)) = c
i
(x

) +
x

(0)
for . Since ((h

c
i
)
i
)(x) = h(x + K) c
i
(x) for x F

, it follows
that h

c
i

i
(x

) : belongs to V [G
i
][G
i+1,
]. Moreover, for < ,

i
(x

) is determined in V [G
i
][G
i+1,
], and hence so are h(x

+ K) =
x

(
i
(x

)) and
(h

c
i
)(
i
(x

)). Therefore h

c
i
belongs to V [G
i
][G
i+1,
] (since it is determined by
its values on
i
(x

) :
1
(1)). Let f = h

c
i
. For each E, there exist
p

G
i+1,
and k

Z such that
p

|
Qi

f(
i
(z
,m
)) = k

.
Since Q
i,
and Z are countable, there exist p G
i+1,
, k Z, and a stationary S V [G
i
]
such that for S, p |
Qi

f(
i
(z
,m
)) = k. Then the (constant) sequence (f(
i
(z
,m
)) :
S) belongs to V [G
i
], so by Corollary 2.5, f belongs to V [G
i
].
So it remains to prove (III.2). Work in V [G
i
]. Let
D
i,
= q Q
i
: (cont(q) )n [(

(n) < ) x

(n)
dom(q)].
Then D
i,
is a dense subset of Q
i
. We claim that it is true in V [G
i
] that:
(III.3) there is a countable ordinal
1
E such that for every y F

,
t, c
1
, c
2
Z, and q
1
, q
2
D
i,
with q
1
= q
2
, if
q

|
Qi

i
(y) = t

h(y +K) = c

for = 1, 2, then c
1
= c
2
.
6
2
1


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
8
-
0
3







m
o
d
i
f
i
e
d
:
1
9
9
9
-
0
8
-
0
3


12 PAUL C. EKLOF AND SAHARON SHELAH
Clearly this implies (III.2). Indeed, we dene
y
(t) to be c if there is a q D
i,
such
that q G
i+1,
and q |
Qi/Gi+1,

i
(y) = t

h(y +K) = c and otherwise
y
(t) = 0.
By (III.3),
y
(t) is well-dened.
proof of (III.3). The proof is by contradiction and uses some of the methods of
the proof of Proposition 2.1. So suppose that for every
1
E there are y

,
t

, c

1
, c

2
Z, and q

1
, q

2
D
i,
such that q

1
= q

2
and q

|
Qi

i
(y

) =
t

(y) = c

where c

1
,= c

2
for = 1, 2. Then, by Fodors Lemma and counting,
there is a p
0
G
i
, t, c
1
, c
2
Z, q V and names

S, q

, y

such that
p
0
|
Pi

S is a stationary subset of
1
E s.t. for all

S,

= t, c

1
= c
1
, c

2
= c
2
and q

1
= q

2
= q
and moreover such that p
0
forces the names to be a counterexample to (III.3), as above.
There is a stationary subset S

1
E such that for every S

there is a condition
p

p
0
in P
i
which forces

S and forces values (elements of V ) to q

and to y

.
Moreover, we can suppose that the p

(i, q

) P
i+1
( = 1, 2) are as in () [cf. proof
of Proposition 2.1] and that p

: S

is as in () [with in place of , but since

/ E, the last sentence does not apply]. Let p

be the heart of p

: S

. We can
also assume that q

: S

forms a -system with heart q (for = 1, 2).

For each E, there is p

P
i+1
and k

Z such that p

i p

, p

(i) q and
p

h(z
,0
) = k

. There is a stationary

S E such that p

:

S satises () and
(); in particular, r = r
p

(0)

for

S and
p

(0)

(n), g
p

(0)
(, n) = g(n) and u
p

(0)
(, n)
are independent of for each n < r. Moreover we can assume that there is k Z such
that k

= k for all

S. Let p

be the heart of p

:

S (so p

(i, q)).
Choose m such that 2
m
does not divide c
1
c
2
. Let M max(g(n) : n < r m)
and let
N = 2
1+( r+1)M(d+1)
(where d is the size of the domain of p

0). Choose

0
< ... <
N1
< <
0
< ... <
N1
where
j
S

,
j


S, every ordinal which occurs in p

is <
0
, and for all j N 1
every ordinal which occurs in p

j
or in q
j

( = 1, 2) is less than
j+1
(where
N
is
taken to be ); and for all j < N 1, every ordinal which occurs in p
j
is less than
j+1
.
Then there is a condition q
0
Q
0
which extends p

(0) and each p

j
(0) and p

j
(0) such
that q
0
forces for all j < N:

j
( r) = ;
j
( r + 1) =
j1
+ 1; u(
j
, r) = y

j
; and g(
j
, r) = 2
m
where
1
= + 1.
As in the proof of Proposition 2.1, there is a condition q

P
i
and a subset W

of N
of size 2
1+( r+1)M
such that q

(0) = q
0
, q

j
for all j N 1, and q

j
i for
j W

. Repeating the argument one more time and using the facts that q
j
1
and q
j
2
force
the same value to (y
j
) and that q
j
1

j
= q

j
2

j
= q, there is a subset W = j, j
o

of W

such that for any function f : W 1, 2 there is a condition q

P
i+1
such that
q

i = q

and q

(i) is an upper bound of p

j
(i), p

jo
(i) q
j
1
, q
jo
2
. In a generic
extension V [G

] where q

we have (since h(z

j,0
+ K) = h(z
jo
,0
+ K) = k and
g(, n) and u(, n) are independent of

S for n < r) that 2
m
divides
h(u(
j
, r)) h(u(
jo
, r)) = h(y

j
+K) h(y

jo
+K) = c
1
c
2
6
2
1


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
8
-
0
3







m
o
d
i
f
i
e
d
:
1
9
9
9
-
0
8
-
0
3


A NON-REFLEXIVE WHITEHEAD GROUP 13
which is a contradiction of the choice of m. This proves (III.3) and thus nally completes
the proof of Theorem 1.5.
5. Theorem 0.2
To prove Theorem 0.2 we use a variant of the iterated forcing that is described in
section 1. Let Q
0
and Q

be as dened there. We shall use a nite support iteration

P

i
,

Q
i
: 0 i <
2

; the

Q
i
are dened inductively. We consider an enumeration,
as before, of names

i
: i <
2
for functions from K to Z. In V
Pi
we dene

Q
i
=

0 if the s.e.s c
i
is torsion
Q

i
otherwise
We claim that if Gis P

-generic, then in V [G] (i) Ext(A, Z) is torsion and (ii) Hom(A, Z) =

0.
To see why (i) holds, consider Hom(K, Z). For some i
2
,

i
is a name for .
In V [G
i
] either represents a torsion element of Ext(A, Z) or else, by construction, in
V [G
i+1
] = [K for some Hom(F, Z), in which case represents the zero element
of Ext(A, Z).
To prove (ii), it suces to show for all i
2
that if h Hom(A, Z)
V [Gi+1]
, then
h Hom(A, Z)
V [Gi]
. If not, then

Q
i
,= 0; but then by the arguments in section 4 it
follows that c
i
is torsion, so

Q
i
= 0, a contradiction.
6. co-Moore spaces
Following [7] we call a topological space X a co-Moore space of type (G, n), where
n 1, if its reduced integral cohomology groups satisfy

H
i
(X, Z) =

G if i = n
0 otherwise.
For n 2, application of the Universal Coecient Theorem shows that
() there exist B
1
and B
2
such that G

= Hom(B
1
, Z)Ext(B
2
, Z) where
Ext(B
1
, Z) = 0 = Hom(B
2
, Z).
Conversely, if G satises (), then there is a co-Moore space of type (G, n) for any
n 2 (cf. [7, Thm. 5], [6]). A sucient condition for G to be of the form () is that
G = DC where C is compact and D is isomorphic to a direct product of copies of Z ([7,
Thm. 5]). In a model of ZFC where every W-group is free, this condition is necessary
(cf. [7, Thm. 3(a)] and [3, Thm. 2.20]); in particular the (torsion-free) rank of C is of
the form 2

for some innite cardinal . However, as a consequence of our proofs we

have:
Corollary 6.1. It is consistent with ZFC + 2
0
= 2
1
=
2
that there is a group A of
cardinality
1
such that Hom(A, Z) = 0 but Ext(A, Z) does not admit a compact topology.
Corollary 6.2. It is consistent with ZFC + 2
0
= 2
1
=
2
that for any n 2 there is
a co-Moore space of type (F, n) where F is the free abelian group of rank
2
.
Corollary 6.3. It is consistent with ZFC + 2
0
= 2
1
=
2
that for any n 2 there is
a co-Moore space of type (C, n) for some uncountable torsion divisible group C.
6
2
1


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
8
-
0
3







m
o
d
i
f
i
e
d
:
1
9
9
9
-
0
8
-
0
3


14 PAUL C. EKLOF AND SAHARON SHELAH
Compare Corollary 6.3 with [6, 2.5 and 2.6]. The conclusions of the corollaries are not
provable in ZFC. Moreover, by an easy modication we can replace
2
in the corollaries
by any regular cardinal greater than
1
. (Note that by [1, Thm. 5.6], Hom(B
1
, Z) cannot
be the free group of rank
1
if Ext(B
1
, Z) = 0.)
References
[1] S. U. Chase, Locally free modules and a problem of Whitehead, Illinois J. Math. 6 (1962), 682699.
[2] S. U. Chase, On group extensions and a problem of J. H. C. Whitehead in Topics in Abelian
Groups, Scott, Foresman and Co., (1963), 173197.
[3] P. C. Eklof and M. Huber, On the rank of Ext, Math. Zeit. 174 (1980), 159185.
[4] P. C. Eklof and A. H. Mekler, Almost Free Modules, North-Holland (1990).
[5] L. Fuchs, Innite Abelian Groups, vols I and II, Academic Press (1970, 1973).
[6] M. Golasi nski and D. L. Goncalves, On co-Moore spaces, Math. Scand. 83 (1998), 4252.
[7] H. Hiller, M. Huber and S. Shelah, The structure of Ext(A, Z) and V = L, Math. Zeit. 162 (1978),
3950.
[8] M. Huber, On reexive modules and abelian groups, J. Algebra 82 (1983), 469487.
[9] T. Jech, Multiple Forcing, Cambridge University Press (1986).
[10] K. Kunen, Set Theory:An Introduction to Independence Proofs, North-Holland (1980).
[11] A. Mekler and S. Shelah, Diamond and -systems, Fund. Math. 131 (1988), 4551.
[12] S. Shelah, Whitehead groups may not be free even assuming CH, I, Israel J. Math. 28 (1977),
193203.
[13] S. Shelah, Whitehead groups may not be free even assuming CH, II, Israel J. Math. 35 (1980),
257285.
(Eklof) Math Dept, UCI, Irvine, CA 92697-3875
E-mail address: peklof@math.uci.edu
(Shelah) Institute of Mathematics, Hebrew University, Jerusalem 91904, Israel
E-mail address: shelah@math.huji.ac.il