(

9
1
4
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
6
-
0
7







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
6
-
0
9


THE FIRST ALMOST FREE WHITEHEAD GROUP
SH914
SAHARON SHELAH
Abstract. Assume G.C.H. and is the rst uncountable cardinal such that
there is a non-free -free abelian Whitehead group of cardinality . We prove
that if all -free Abelian group of cardinality are Whitehead then is nec-
essarily an inaccessible cardinal.
0. Introduction
For the non-specialist reader, note that we deal exclusively with abelian groups,
we call such G Whitehead when : if Z H and H/Z

= G then Z is a direct
summand of H. Why this property is worthwhile and generally on the subject see
the book of Eklof-Mekler [EM02]. Recall G is a free abelian group if it is the direct
sum of copies of (Z, +). All free abelian groups are Whitehead and every Whitehead
group is somewhat free (
1
-free, see below). Possibly (i.e. consistently with ZFC)
every Whitehead group is free and possibly not (in fact, it seems that almost any
behaviour is possible).
Recall that G.C.H., the generalized continuum hypothesis, says that 2

=
+
for
every innite and is an inaccessible cardinal means that: it is strong limit (i.e.
< 2

< ) and regular, i.e. >

i<

i
: when < and i <
i
< .
It is well known that inaccessible cardinals are large, e.g. it is not provable in
ZFC, the usual axioms of set theory that such cardinals exists.
Also being -free of cardinality is a central notion where we say an abelian
group G is -free when the pure closure inside G of any subgroup generated by
< elements is free. This explains the interest in the result, which restricts the
behaviour. That is, assume is the rst cardinal such that there is a non-free
-free abelian Whitehead group of cardinality . The conclusion is, assuming GCH
and not strongly inaccessible, that not all such abelian groups are Whitehead.
Set theorists may recall that in [Sh:667, 1] it is proved that if is strong
limit singular, =
+
= 2

and S < : cf() = cf() is stationary,

then though
S
may (= consistently) fail, still we can prove a relative of
S
sucient for constructing abelian groups G of cardinality related to satisfying
Hom(G, Z) ,= 0. We prove here in 1.5 somewhat more and use it in the proof
of the Theorem 1.1. This claim is complementary to [Sh:587] which shows that
consistently there is such regular (in fact strongly inaccessible) .
Date: June 7, 2011.
1991 Mathematics Subject Classication. [2010] Primary 03E75; Secondary: 03C60, 20K20.
Key words and phrases. Abelian group, Whitehead group, almost free, stationary sets, -sets.
Research supported by German-Israeli Foundation for Scientic Research/Development Grant
No. I-706-54.6/2001. I would like to thank Alice Leonhardt for the beautiful typing. First Typed
- 05/Nov/17.
1
(
9
1
4
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
6
-
0
7







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
6
-
0
9


2 SAHARON SHELAH
{5n.0A}
Convention 0.1. Saying group we mean abelian group.
In [Sh:587] we answer
{5n.0D}
Question 0.2. (G obel) Assuming GCH can there be a regular such that:

(a) = cf() >

0
and there are -free not free groups of cardinality
(b) every -free group of cardinality is a Whitehead group.
Moreover by [Sh:587] it is consistent that: G.C.H. + for some strongly inaccessible
we have

where the statement

is dened by:

(a) is regular uncountable and there are -free non-free (abelian) groups
of cardinality
(b) every -free (abelian) group of cardinality is a Whitehead group
(c) every Whitehead group of cardinality < is free.
[For clause (c), the following sucient condition is used there: for every regular
uncountable < we have

. The proof starts with weakly compact and adds

no new sequence of length < , so e.g. starting there with L this holds.]
A natural question is whether
{5n.0H}
Question 0.3. Assume G.C.H., is it consistent that there is an accessible such
that

holds? In other words there is cardinal satisfying the following and there
rst such is accessible:
() there is a -free non-free Whitehead group of cardinality .
Now Theorem 1.1 says that no. But another natural question is:
{5n.0K}
Question 0.4. Assume G.C.H. Can there be a such that

but is the rst such

accessible?
This is problem F4 of [EM02] and it remains open.
{0z.11}
Denition 0.5. 1) An abelian group G is free if G is the direct sum of copies of
Z.
2) For a cardinality
1
, an abelian group G is -free when every subgroup of
G of cardinality < is free.
3)

G = G

: < ) is a ltration of the abelian group G of cardinality if G

is
a subgroup of G of cardinality < , increasing continuous with and G = G

:
< .
4) G, an abelian group of cardinality , is strongly -free when it is -free and for
every subgroup G

of G of cardinality < there is a subgroup G

of cardinality
< such that G

and G/G

is -free.
We thank the referees for many helpful comments, in particular for making the
work more self-cointained.
(
9
1
4
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
6
-
0
7







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
6
-
0
9


THE FIRST ALMOST FREE WHITEHEAD GROUP SH914 3
1. The First almost free non-free Whitehead
{5n.1}
Theorem 1.1. (G.C.H.) Let be the rst >
0
such that there is a -free abelian
Whitehead group, not free, of cardinality (and we assume there is such ). If
is not (strongly) inaccessible, then there is a non-Whitehead group G of cardinality
which is -free (and necessarily non-free).
Proof. Stage A: Let G exemplify the choice of .
Necessarily is regular, by the singular compactness, see e.g. [EM02, Ch.IV,3.5].
Let

G = G

: < ) be a ltration of G so as G is -free necessarily every

G

is free. Without loss of generality each G

is a pure subgroup of G. Let

S := (

G) = < : is a limit ordinal and G/G

is not -free. Now recall G

is not free hence (

G) is stationary.
Stage B: G is strongly -free.
Toward contradiction assume that not. Without loss of generality is a successor
cardinal (why? by the theorems assumption and being regular but also as for
a limit cardinal, -free implies strongly -free).
By our present assumption toward contradiction, for some < for every
[, ), the abelian group, G/G

is not -free. Hence without loss of generality

< G
+1
/G

is not free and if G

+1
/G

is uncountable then for some

,
it is

-free not free. By the theorem assumption (and as countable Whitehead

groups are free), < G
+1
/G

is not a Whitehead group, i.e. (

G) = . But
is a successor cardinal, so let =
+
; also 2

< 2

as we are assuming GCH so

the weak diamond holds for (see [DvSh:65]) so by the previous sentence and e.g.
[EM02, 1.10,pg.369], we know that G is not a Whitehead group, contradiction to
the assumption on G.
Stage C: Hence without loss of generality
()
1
if is a non-limit ordinal then G/G

is -free
and obviously without loss of generality
()
2
(a) if S then G
+1
/G

is not free
(b) if < then G

is a pure subgroup of G and is free

hence G/G

is torsion free.
Also we can choose

H = H

: S) such that
()
3
(a) H

is a subgroup of G
(b) H

/(H

) is not free
(c) under (a) + (b) the rank of H

/(H

) is minimal call it

hence

is <
0
or is regular uncountable <
(d) the cardinality of H

is

+
0
, in fact equality holds.
Note that [H

[ may be < [G

[ so it is unreasonable to ask for G

.
()
4
without loss of generality
(a) H

G
+1
(b) G

/(H

) is free
(c) G

+ H

is a pure subgroup of G
+1
.
(
9
1
4
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
6
-
0
7







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
6
-
0
9


4 SAHARON SHELAH
[Why? For clause (a) we can restrict

G to a club. For clause (c) let H

be the
pure closure of H

+ G

inside G
+1
so H

/G

, (H

+ G

)/G

, H

/(H

)
has the same rank, which is

. As G
+1
/G

is torsion free, also H

/G

is torsion
free. Hence there is H

of cardinality

+
0
such that G

+ H

is a
pure subgroup of H

hence of G
+1
and H

+ G

= H

and e.g. the rank of

H

/(H

) is the same as the rank of H

/G

which is

, so replacing H

by
H

also clause (c) holds.

For clause (b) note that G

is free; so there is G

of cardinality

+
0
such that G

/G

is free and H

and replace H

by H

= H

+ G

noting that H

+ G

= H

+ G

.]
By the hypothesis on , if K is a -free not free group of cardinality and

0
< < then K is not a Whitehead group so clearly (recalling that Whitehead
groups which are countable are free):
()
5
H

/(G

) is not Whitehead for S.

Hence by [Sh:44] or see [EM02, Ch.VI,1.13] we know that
()
6

S
.
Recall is regular uncountable so toward contradiction assume =
+
. Let
= cf(). But, [Sh:108] or [EM02, Ch.VI,1.13], GCH implies that
S
for ev-
ery stationary S

where S

:= < : cf() = .
Hence we know that
()
7
for some club E of we have S E S

so without loss of generality

S S

, i.e., S cf() = .
Also without loss of generality (from strongly -free)
()
8
S

+
0
.
[Why? For completeness we elaborate. Let S
1
= S :

+
0
< ; rst assume
S
1
is stationary. For each S
1
necessarily H

has cardinality

+
0
< =
cf() hence for some

< , H

. By Fodor lemma for some () <

the set S
2
:= S
1
:

= () is stationary. As = cf(), =
+
, clearly

<
= recalling GCH hence H

G
()
: S
2
has cardinality hence
S
3
= S
2
: H

G
()
= H

is a stationary subset of for some

and H

H
()
. Let

be the -th member of S

3
and H

= H

+H

: <
for (

+
0
)
+
. Clearly for = (

+
0
)
+
, H

is not free and has cardinality

= cf() < , contradiction to G is -free. So necessarily S
1
is not stationary,
and as we can restrict

G to a club, without loss of generality S
1
= so ()
8
holds
indeed.]
Stage D: S

< .
Why? Otherwise for some S,

= so

is innite hence

is regular
and by ()
3
uncountable and there is a

-free not free group of cardinality

hence is regular and there is a -free not free abelian group of cardinality < ,
i.e. H

/(G

), hence this group is not Whitehead. So there is a sequence

H

: + 1) purely increasing continuous sequence of free groups such that

< implies H

+1
/H

is free but H

+1
/H

is not free and is isomorphic to

H

/(G

).
(
9
1
4
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
6
-
0
7







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
6
-
0
9


THE FIRST ALMOST FREE WHITEHEAD GROUP SH914 5
[Why? Let y

: < ) be a free basis of G

, so G

= Zy

: < . As
y

: < G

and cf() = , there is an increasing continuous sequence

: < ) of ordinals < with limit such that y

G
+1
.
Let H

= Zy

: < for < , H

= G

, H

+1
= G

+ H

, they are as
required. E.g. why is H

+1
/H

free? as G
+1
/G
+1
is free by ()
1
, also G

/H

is
free as y

+H

: [, ) is a free basis but H

G
+1
G

hence G
+1
/H

is free so together G
+1
/H

is free which implies that its subgroup H

+1
/H

is
free as promised.]
We can nd H

and H

, h

:

2) such that:
(a) H

= Zx
t
: t I and [I[ =
(b) I is the disjoint union of I

:
>
2
(c) [I

[ = rk(H

g()+1
/H

g()
) for
>
2
(d) H

= Zx
t
: t I

: < g() H

for
>
2
(e) h

is an isomorphism from H
g()
onto H

(f) h

if < g(),
>
2.
Now we can nd H
+

, h
+

:

2) such that
(g) H

H
+

(h) h
+

is an isomorphism from H

+1
onto H
+

extending h

.
Without loss of generality
(i) H
+

= H

so there is
(j) H

extends H
+

and H

such that H
+

generates H

and H

/H

is
isomorphic to H

/H

.
Lastly, without loss of generality
(k) the sets H

:

2) are pairwise disjoint,
so there is an abelian group H such that
(l) H

H for

2 and
(m) H/H

= H

/H

:

2
(n) H has cardinality 2

= .
Next we note
(o) H is -free.
[Why? Note that H
+

:

2 include x
t
: t I hence the subgroup it
generates includes H. Let H

H be a sub-group of cardinality < so , hence

there are
i


2 for i < such that H

H
+
i
: i < and j < i
i
,=
j
. We
can choose (i) < by induction on i < such that
i
: [(i), ) : i <
is a sequence of pairwise disjoint sets.
For each i let H

i
H
+
i
be such that H
+
i
= H

i
H
i(i)
. Let us dene H

i
for i by: H

0
H is generated by x
t
: t I but t / I
i
: i < and
[(i), ) and H

i
is H

j
: j < i H

0
. Clearly H

H, H

i
: i )
(
9
1
4
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
6
-
0
7







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
6
-
0
9


6 SAHARON SHELAH
is increasing continuous, H

0
is free and H

i+1
/H

= H

i
is free. It follows that H

is free hence H

is free. So clause (o) holds indeed.]

Let

: < ) list

2, let H

H be H
+

: < +H

,

H

= H

: <
) is a ltration of H, and (

H

) = and H

+1
/H

+1
is isomorphic to H

+1
/H

so is not free and is not Whitehead. Hence, see [EM02, Ch.XII,1.10,pg.369], H is

not Whitehead. So H is -free (see clause (o) of cardinality (see clause (o) and
not Whitehead, and so not free), the desired conclusion of the theorem. So indeed
without loss of generality the stage desired conclusion, S

< holds.]
Stage E: is singular.
Why? Because by earlier stages cf() = and S implies cf()

<
and S = cf(), so necessarily is singular.
Stage F: Let = cf() so is regular < ; also choose = cf() < such that
S

1
:= S :

= is stationary. Note that =

<
as G.C.H. holds recalling
= cf() and S cf() = by ()
7
. We shall now use [Sh:521, 3] and its
notation, see 1.2, 1.3, 1.5 below.
By the Theorem 1.2 below we can nd a -witness x, so it consists of n
1, S, B

: S
c
), s

: S
f
, < n) as there, i.e. as in [Sh:521, 3.6] with
(,
+
, S) there standing for (,
1
, S) here, such that
() ) S S

1
, this means W(<>, S) = S

1
.
Clearly continuing to use the notation there, S

1
(), S) hence being
regular it is < , and so without loss of generality constant (in fact it is by the
proof). Now we apply the claim 1.5 below with there standing for here.
Why clause (d) of the assumption of 1.5 holds? If >
0
, by [Sh:521, 1.2], the
group G
x(<>)
, derived from x()) is a (< , S
x
)-free non-free abelian group
of cardinality (), S
x
) = , so is not Whitehead by the theorem assumption on
; note that G
x(<>)
was derived in some way from H

/(H

), but it is not
necessarily equal to it. If
0
then G
x(<>)
is a non-free (abelian) countable
group hence is not Whitehead.
So the assumption of 1.5 says that there is a strongly -free abelian group G
of cardinality by the theorems assumption which is not Whitehead, so we are
done.
1.1
Recall (by [Sh:161, 5] and see [Sh:521, 3]; or see [EM02]).
{1n.4}
Theorem 1.2. For any >
0
the following conditions are equivalent:
(a) there is a -free not free abelian group
(b) PT(,
1
) which means that: there is a family P of countable sets of car-
dinality with no transversal (i.e. a one-to-one choice function) but any
subfamily of cardinality < has a transversal
(c) there is a -witness x as in [Sh:521, 3.6,3.7]; so x consists of
() a natural number n
) a so-called -set S
n
: decreasing closed under initial
segments, (see [Sh:521, 3.1])
() disjoint -system

B = B

: S
c
), see [Sh:521, 3.4]
() s = s

: S
f
, < n), etc.
(
9
1
4
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
6
-
0
7







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
6
-
0
9


THE FIRST ALMOST FREE WHITEHEAD GROUP SH914 7
(d) Without loss of generality B

: S
c
= , so S = S
x
, etc.
(e) Let a

,m
: m < ) list s
,x

with no repetition for S

x
f
, < n and
a
(1)
(1),m(1)
= a
(2)
(2),m(2)
implies (1) = (2), m(1) = m(2) and m < m(1)
a
(1)
(),m
= a
(2)
(2),m
, so without loss of generality a

,m
< a

,m+1
for every
relevant , , m; also = sup s
0,x

: ) S
f
when ) S
f
and
(, S
x
), W(, S
x
) are well dened.
Used but not named in [Sh:521]:
{1n.6}
Denition/Claim 1.3. 1) For a -witness x and S
x
we dene x

= x() by
n
x()
= n
x
g(), S
x()
= : S
x
, B
x()

= B
x

, s
,x()

= s
g()+,x

.
2) For a -witness x let G
x
be the abelian group G
{<>:W(<>,Sx)}
dened inside
the proof of [Sh:521, 1.2].
{a8}
Remark 1.4. We may use the following.
1) For a -witness x let
K
x
= I S
x
: I is a set of pairwise -incomparable sequences such that
: ) I is an initial segment of
W(, S) for any S.
2) For x a -witness and I K
x
let Y [I] and G
I
be dened as in the proof of
[Sh:521, 1.2] before Fact A. We may write instead of I = .
{5n.2}
Claim 1.5. Assume
(a) is strong limit singular, =
+
= 2

and = cf()
(b) S < : cf() = is stationary
(c) x is a -witness see 1.2 with W(), S) S
(d) for each W(), S), the abelian group G
x(<>)
is not Whitehead (where
G
x(<>)
is dened as inside the proof of [Sh:521, 1.2]).
Then
1) There is a strongly -free abelian group G of cardinality which is not Whitehead,
in fact (G) S.
2) There is a strongly -free abelian group G

of cardinality satisfying HOM(G

, Z) =
0, in fact (G

) S (in fact the same abelian group can serve).

Remark 1.6. 1) We rely on [Sh:521, 3].
2) So in clause (d), G
x(<>)
is the abelian group dened from x()).
3) If you do not like clause (d) of 1.5, replace it by is as in 1.1.
Proof. 1) Let S = S
x
, etc. Without loss of generality
()
0
B
x

: S
c
= .
Let H () =

<
M

where M

(H (), ) has cardinality , is increasing con-

tinuous with such that + 1 M
0
and M

: ) M
+1
.
Let S
0
= W(), S) : M

= , as W(), S) is stationary and < :

M

= is a club of ; clearly also S

0
is a stationary subset of ; now for each
S
0
as is strong limit (and so =
<
) clearly
(
9
1
4
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
6
-
0
7







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
6
-
0
9


8 SAHARON SHELAH
()
1
if Y M

, [Y [ < and cf() = then there is an increasing continuous

sequence X
i
: i < ) of subsets of Y with union Y such that M

X
i
:
i < .
As W(), S) (), S) < clearly for some
()
2
is regular < the set S
1
= S
0
: (), S) = is stationary.
For each S
1
clearly x()) is a -witness and let
()
3
(a) X

:= (, s) : S
x(<>)
f
(i.e. is -maximal in S
x(<>)
) and s
is a nite initial segment of s
0

which is a set of members of

B

of order type .
Clearly
()
4
X

has cardinality .
We can nd

X

such that
()
5

X

= X
,i
: i < ) is as in ()
1
above, i.e. is -increasing continuous,
so X
,i
M

, [X
,i
[ and letting X
,
:= X
,i
: i < we have
X
,
= X

()
6
Z
,i
:= (, [s[) : (, s) X
,i
for i .
We dene equivalence relation E on S
1
:
()
7

1
E
2
i
(a) S
1
= S
2
equivalently S
x(1)
= S
x(2)
(b) for each i < we have Z
1,i
= Z
2,i
Clearly E is actually an equivalence relation but < and is strong limit hence
E has 2

< < equivalence classes. So for some

S
1
, S
2
:=

/E is a
stationary subset of . Clearly there is F M
0
which is a one to one function with
range : < , = is < but > 0 and with domain H () hence for every
S
2
it maps M

onto Rang(F) and without loss of generality if

1

2
are
from
>
H () then F(
1
) < F(
2
) and F(X
,j
: j i)) is > sups Rang() :
(, s) X
,i
.
Let

,i
= F(X
,j
: j i)) for S
2
, i < , so clearly

,i
: i < ) is increasing
with limit by the choice of F and of S
2
, as (see 1.2(2) - = sups
0,x

: )
S
f
). By [Sh:667, 4.2], we can nd (
,1
,
,2
) : S
2
) such that
(a)
,

(b)
,1
(i) <
,2
(i) for i <
(c) if c : 2

+ then for stationarily many S

2
we have
i < < c(

,i
+ +
,1
(i)) = c(

,i
+ +
,2
(i)).
Let y = xS

where S

= ) S : ,= ) and (0) S
2
.
Now at last we shall dene the group. Essentially it will be similar to the
group G
x
, see 1.3(2) so dened inside the proof of [Sh:521, 1.2] from the system x,
restricted to S

only, but whereas in ()

a
I,
before Fact A in the proof in [Sh:521,
1.2] we use 2y
,m+1
= y
,m
+ x[a

,m
) : < n and a

,m
Y [I] here we replace
x[a

,m
] by the dierence of two, related to ; this may become clearer after reading
(
9
1
4
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
6
-
0
7







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
6
-
0
9


THE FIRST ALMOST FREE WHITEHEAD GROUP SH914 9
the proof. The -freeness will be inherited from S being -free. The non-Whitehead
comes from .
For S
1
let g

: S
x()
be a one-to-one function, so by the choice of S
2
for some g we have S
2
g

= g.
As in [Sh:521, 1] for each S
f
and < n recall, a

,m
: m < ) lists
s

B
(+1)
, let Y = B

: S
c
B
x

and for S
f
, m < let i
m
() be
the minimal i such that ( [1, n), a
0
,
: m) X
(0),i
. We dene G as the
abelian group generated by
= y
,m
: m < and S

f
x[a] : a Y z

: <
freely except for the equations
() for S

f
and m < , so := (0) S
2
the equation (letting i = i
m
())
2y
,m+1
= y
,m
+x[a

,m
] : 0 < < n
+z

,i
+g()+
,2
(i)
z

,i
+g()+
,1
(i)
recalling g : S
f
such that g S
f
: (0) = is one to one,

,i
= F(X
,j
: j i)).
For let G

be the subgroup of G generated by

y
,m
: m < and S

f
and (0) <
x[a] : a Y and a + 1 <
z

: < moreover + 1 < .

Easily

1
G

is a pure subgroup of G, increasing continuous with

2
if S
2
then G
+1
/G

is isomorphic to G
x()
which is not Whitehead.
[Why? By clause (d) of the assumption; now at rst glance the set of generators
and equations in the proof of [Sh:521, 1.2] and in this proof are dierent. But note
that only y

, ) S
f
and x[a], a s

: [1, n), ) S
f
appear in
the equation. Alternatively use Remark 1.4 and prove G
+1
/G

is isomorphic to
G
x,+1
/G
x,
in [Sh:521, 1.2] notation; again note that the z

- here and x[a], a

s
0,x

: S
f
disappear.]
But recall

3
G is strongly -free, moreover if S
2
and (, ) then G

/G

is
free.
[Why? As in the proof of Fact A inside the proof of [Sh:521, 1.2].]

4
G is not Whitehead.
[Why? We choose (H

, h

, g

) by induction on such that

(a) H

is an abelian group extending Z

(b) h

is a homomorphism from H

onto G

with kernel Z
(
9
1
4
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
6
-
0
7







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
6
-
0
9


10 SAHARON SHELAH
(c) g

is a function from G

to H

inverting h

(but in general not a homo-

morphism)
(d) H

is increasing continuous with

(e) h

is increasing continuous with

(f) g

is increasing continuous with

(g) if = + 1 and S
2
then there is no homomorphism g

from G

into
H

inverting h

such that:
i < < g

(z

,i
++
,1
(i)
)g
+1
(z

,i
++
,1
) = g

(z

,i
++
,2
(i)
)
g
+1
(z

,i
++
,2
(i)
)
(note: the subtraction in Z, the kernel of h

)
For = 0, limit and = +1, / S
2
this is obvious. For = +1, S
2
it
is known that if instead of in clause (g) we know g

this is possible. But

gives all the necessary information. In more details let G

be the subgroup of G

generated by z

,i
++
,2
(i)
z

,i
++
,1
(i)
: < and i < .
Let G

+1
be the subgroup of G
+1
generated by G

y
,m
: ) S

f
.
Clearly G

is a pure subgroup of G

+1
and of G

and G
+1
= G

+1

.
Let H

= h
1

(G

), clearly h

is a homomorphism from H

onto G

with
kernel Z.
Clearly

4.1
if g

, g

Hom(G

, H
+1
) invert h

and both satises of clause (g) then

g

= g

.
So [G

[ 1 where
G

= gG

: g is a homomorphism from G

to H

inverting h

and satisfying in clause (g).

Let g

be the unique member of G if G is non-empty and otherwise let it be any

homomorphism from G

into H

inverting h

, exist as G

is free.
We now choose (H

+1
, h

+1
) such that
H

+1
h

+1
Hom(H

+1
, G

+1
)
h

+1
has kernel Z
h

+1
extends h

cannot be extended.
Next without loss of generality H

+1
H

= H

, let H
+1
= H

+1

H

, and let
h
+1
Hom(H
+1
, G
+1
) extend h

, h

+1
. Let g
+1
g

invert h
+1
but is not
necessarily a homomorphism. So we have chosen (H

, h

, g

) such that: if G

,=
then we cannot nd g

Hom(G

+1
, Z) inverting h

such that g

. This
suces for carrying out the induction.
Having carried the induction, clearly h = h

is a homomorphism from H = H

onto G

= G with kernel Z. To show that G is not a Whitehead group it suces to

prove that h is not invertible as a homomorphism. But if g Hom(G, H) inverts
h then x G g(x) g

(x) Z.
(
9
1
4
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
6
-
0
7







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
6
-
0
9


THE FIRST ALMOST FREE WHITEHEAD GROUP SH914 11
We dene a function f with domain : for < , < : f ( + ) =
g(z

2
++
) g

(z

2
++
) : < ).
So for stationarily many S
2
we have i < f (

,i
+
,1
(i)) = f (

,i
+

,2
(i)). For any such we get a contradiction by clause (g) of the construction, so
we have proved
4
.
This nishes the proof of part (1), as G = G

is as required.
2) For the proof of part (2) can use:
for regular uncountable , the following conditions are equivalent
(a) every -free abelian group of cardinality is Whitehead
(b) for every -free abelian group of cardinality we have Hom(G, Z) ,= 0.
[Why? If (a) and G as in(b), let h be a pure embedding of Z into G, let G

=
G/Rang(h) and use the denition of G

is a Whitehead group. If G, H and h

Hom(H, G) form a counterexample then we can nd a purely increasing continuous
sequence G

: ) and x

: < ) such that: x

: < = x G

: Zx
is a pure subgroup of G

and for each , there is a pure embedding h

of H into
G
+1
such that h(1
Z
) = x

and G
+1
= G

Zx
h

(H).
Easily G

contradicts clause (b).]

We can also construct directly.
1.5
(
9
1
4
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
6
-
0
7







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
6
-
0
9


12 SAHARON SHELAH
2. Private Appendix
Assignments
1) The [EM02] conjecture: maybe can be contradicted in [Sh:832].
2) Assume G.C.H. and is rst inaccessible but every stationary set reects and
more every -free abelian group is free. Prove the proofs in [Sh:587] apply.
(
9
1
4
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
6
-
0
7







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
6
-
0
9


THE FIRST ALMOST FREE WHITEHEAD GROUP SH914 13
3. Private Appendix
{0n.21}
Discussion 3.1. Toward solving [?, Problem F4].
Step A: We start with V [= GCH + diamond on every stationary set, Fsp = :
no almost free group in large enough.
Step B: Choose regular > , force S
1
S

not reecting, and force uniformiza-

tion on S
1
let
S
for every stationary S S
2
, S
1
, S
2
stationary. Still ()

.
Stage D: Let > , force by initial segment =
,i
: S
2
, i S) such that
(a) S
3
S

stationary
(b)
,i
) is increasing of length
(c) sup Rang(
,i
) : S) increases with limit , S
3
, i < j i S
1
j
S
2
sup(Rang(
,i
) <
,j
(0)
(d) every initial segment of is free
(e) on , S
3
reection as of old.
Alternative forcing: S
3
= S

, conditions apr? < = cf().

Stage E: It is enough to prove that: if G is almost free in
+
then (G) S
3
moreover G

: <
+
) a representation then for almost all S
2
, G
+1
/G

is.
References
[EM02] Paul C. Eklof and Alan Mekler, Almost free modules: Set theoretic methods, North
Holland Mathematical Library, vol. 65, NorthHolland Publishing Co., Amsterdam, 2002, Re-
vised Edition.
[Sh:44] Saharon Shelah, Innite abelian groups, Whitehead problem and some constructions, Is-
rael Journal of Mathematics 18 (1974), 243256.
[DvSh:65] Keith J. Devlin and Saharon Shelah, A weak version of which follows from 2

0
<
2

1
, Israel Journal of Mathematics 29 (1978), 239247.
[Sh:108] Saharon Shelah, On successors of singular cardinals, Logic Colloquium 78 (Mons, 1978),
Stud. Logic Foundations Math, vol. 97, North-Holland, Amsterdam-New York, 1979, pp. 357
380.
[Sh:161] , Incompactness in regular cardinals, Notre Dame Journal of Formal Logic 26
(1985), 195228.
[Sh:521] , If there is an exactly -free abelian group then there is an exactly -separable
one, Journal of Symbolic Logic 61 (1996), 12611278, math.LO/9503226.
[Sh:587] , Not collapsing cardinals in (< )support iterations, Israel Journal of
Mathematics 136 (2003), 29115, math.LO/9707225.
[Sh:667] , Successor of singulars: combinatorics and not collapsing cardinals in (< )-
support iterations, Israel Journal of Mathematics 134 (2003), 127155, math.LO/9808140.
[Sh:832] , Incompactness in singular cardinals, Preprint.
Einstein Institute of Mathematics, Edmond J. Safra Campus, Givat Ram, The He-
brew University of Jerusalem, Jerusalem, 91904, Israel, and, Department of Mathe-
matics, Hill Center - Busch Campus, Rutgers, The State University of New Jersey, 110
Frelinghuysen Road, Piscataway, NJ 08854-8019 USA
E-mail address: shelah@math.huji.ac.il
URL: http://shelah.logic.at