4
0
2
)
r
e
v
i
s
i
o
n
:
1
9
9
9

0
8

1
2
m
o
d
i
f
i
e
d
:
2
0
0
3

0
3

0
7
BOREL WHITEHEAD GROUPS
Saharon Shelah
Institute of Mathematics
The Hebrew University
Jerusalem, Israel
Rutgers University
Mathematics Department
New Brunswick, NJ USA
MSRI
Berkeley, CA USA
Abstract. We investigate the Whiteheadness of Borel abelian groups (
1
free,
without loss of generalityas otherwise this is trivial). We show that CH (and even
WCH) implies any such abelian group is free, and always
2
free.
1998 Mathematics Subject Classication 03C60, 03E15, 20K20.
Key words and phrases. Abelian groups, Whitehead groups, Borel sets, Borel Abelian groups.
I would like to thank Alice Leonhardt for the beautiful typing.
First Typed  98/Mar/4
Latest Revision  99/May/17
1, 2 done, Fall 89
Publication no. 402
Typeset by A
M
ST
E
X
1
(
4
0
2
)
r
e
v
i
s
i
o
n
:
1
9
9
9

0
8

1
2
m
o
d
i
f
i
e
d
:
2
0
0
3

0
3

0
7
2 SAHARON SHELAH
0 Introduction
It is independent of set theory whether every Whitehead group is free [Sh 44].
The problem is called Whiteheads problem. In addition, Whiteheads problem
is independent of set theory even under the continuum hypothesis [Sh:98]. An
interesting problem suggested by Dave Marker is the Borel version of Whiteheads
problem, namely,
Question: Is every Whitehead group coded by a Borel set free? (For a precise
denition of a Borel code, see below.) In the present paper, we will give a partial
answer to this question.
0.1 Denition. 1) We say that
=
0
,
1
) is a code for a Borel abelian group
if:
(a)
0
(. . . , . . . ) codes a Borel equivalence relation E = E
on a subset B
= B
of
2 so [
0
(, ) B
] and [
0
(, ) B
& B
], the group
will have a set of elements B = B
/E
(b)
1
=
1
(x, y, z) codes a Borel set of triples from
2 such that
(x/E
, y/E
, z/E
) :
1
(x, y, z) is the graph of a function from B B
to B such that (B, +) is an abelian group.
2) We say Borel
+
if (b) is replaced by:
(b)
1
codes a Borel function from B
to B
which respects E
, the
function is called + and (B, +) is an abelian group (well, we should denote
the function which + induces from (B
/E
) (B
/E
) into B
/E
by e.g.
+
E
, but are not strict).
3) We let B
= B
.
4) An abelian group B is called Borel if it has a Borel code similarly Borel
+
.
Clearly
0.2 Observation: The set of codes for Borel abelian groups is
1
2
.
An interesting problem suggested by Dave Marker is the Borel version of White
heads problem: namely
0.3 Question: Is every Borel
+
Whitehead group free?
In this paper we will give a partial answer to this question, even for the Borel
(without +) version. We will show that every Borel Whitehead group is
2
free. In
particular, the continuum hypothesis implies that every Borel Whitehead group is
free. This latter result provides a contrast to the authors proof ([Sh:98]) that it is
consistent with CH that there is a Whitehead group of cardinality
1
which is not
free.
We refer the reader to [EM] for the necessary background material on abelian
groups.
Suppose B is an uncountable
1
free abelian group. Let S
0
= G B : [G[ =
0
and B/G is not
1
free. It is well known that if B is not
2
free, then S
0
is
(
4
0
2
)
r
e
v
i
s
i
o
n
:
1
9
9
9

0
8

1
2
m
o
d
i
f
i
e
d
:
2
0
0
3

0
3

0
7
BOREL WHITEHEAD GROUPS 3
stationary. We will argue that the converse is true for Borel abelian groups and the
answer is quite absolute. Lastly, we deal with weakening Borel to Souslin.
0.4 Question: If B is an
2
free Borel abelian group, what can be the n in the
analysis of a nonfree
2
free abelian subgroup of B from [Sh 161] (or see [EM] or
[Sh 523])?
We thank Todd Eisworth and the referee for corrections.
(
4
0
2
)
r
e
v
i
s
i
o
n
:
1
9
9
9

0
8

1
2
m
o
d
i
f
i
e
d
:
2
0
0
3

0
3

0
7
4 SAHARON SHELAH
1 On
2
freeness
1.1 Hypothesis. Let B be an
1
free Borel abelian group. Let
be a Borel code
for B.
Let S
B
= S
= B
V
= S
V [G]
,
where
B = B
: <
1
) is an increasing continuous chain of countable pure
subgroups. Let S = <
1
: B/B
is not
1
free. Since S
is stationary (as a
subset of [B]
0
), clearly S is a stationary subset of
1
. So V [G] [= B is not free.
By Pontryagins criteria for each S there are n
and a
0
, . . . , a
such
that
PC(B
0
, . . . , a
)/B
is not free, where PC(X) = PC(X, B) is the pure closure of the subgroup of B
which X generates. We choose n
be a well ordering
of H (). Let N _ (H (), , <
: <
1
) and
0
, . . . , a
) : <
1
_
belong to N.
The model N has been built in V [G], but since forcing with P adds no new reals,
there is a transitive model N
0
V isomorphic to N and let h be an isomorphism
from N onto N
0
. Clearly h maps
to
. From now on we work in V . Hence H ()
below is dierent from the one above.
We build an increasing continuous elementary chain N
: <
1
), choosing
N
by induction on , each N
s are neither
necessarily transitive nor even well founded.
(
4
0
2
)
r
e
v
i
s
i
o
n
:
1
9
9
9

0
8

1
2
m
o
d
i
f
i
e
d
:
2
0
0
3

0
3

0
7
BOREL WHITEHEAD GROUPS 5
Let =
= (v) : N
where
in the vocabulary
, <
i N
be a subset of
such that:
(a) t
and t
then t
(b) t
is directed
(c) for some countable M
(H (), , <
) to which N
belongs, if
M
is a dense subset of
then t
,= .
Clearly by the density if
and
, then
or
.
Thus, t
. Since N
where
N
N
+1
, b
N
+1
realizes t
.
We claim that N
+1
has no new natural numbers, i.e. if N
+1
[= c is a
natural numbers then c N
. Why? As c N
+1
clearly for some f N
we have N
) = c. Let
D
f
=
_
(v)
:N
we have
N
[= (x)((x) f(x) = d)
_
.
It is easy to check that D
f
is a subset of
, it belongs to M
and it is a dense
subset of
; hence t
D
f
,= . Let (x) D
f
t
, so N
+1
[= [b
], and by the
denition of D
f
we get the desired conclusion.
If N
& b
is a countable
ordinal. Also N
+1
[= b
h(S).
We claim that b
in the sense of N
+1
. Assume
N
+1
[= c is a countable ordinal, c < b
so for some f N
we have N
[= f :
1
1
is a function and N
+1
[= c = f(b
), N
+1
[= f(b
) < b
. Then
N
:
N
[= ( <
1
)(v)((v)
f(v) = ) (v)((v) f(v) v). By Fodors lemma (which N
satises) D is
a dense subset of
and clearly D M
. Since t
such that N
+1
[= f(b
) = .
Now N
[= a implies a = n < : N
[= n a so as h(
) =
clearly N
[= x/E
B x/E
B, and N
[= x, y, z B
, x/E
+y/E
=
z/E
x/E
+y/E
= z/E
. Also if N
[= x/E
, y/E
y, then x/E
,= y/E
.
For each <
1
, if N
[= b <
1
, let B
b
be the group (h(
B))
b
as interpreted
in N
, i.e. N
thinks that B
b
is the bth group in the increasing chain h(
B).
Clearly B
b
B if E
b
map (x/E
)
N
to x/E
, so
j
b
embeds B
b
into B; let this image be called G
b
. Also in N
there is a bijection
between B
b
and . If > , since N
_ N
b
= B
b
when E
is equality or j
b
= j
b
and G
b
= G
b
in the general case.
In particular, G
+1
b
is the union of G
b
: N
[= b <
1
.
For <
1
, let G
= G
+1
b
: n
) : S)))(b
) N
+1
be
(a
b
/E
)
N
: m
), so N
+1
thinks that a
b
/E
: m
) witness that
h(B)/B
+1
b
0
/E
, . . . , a
b
/E
G
+1
and
PC(G
a
b
0
/E
, . . . , a
b
/E
)/G
is not free. So G
+1
/G
s but we have to
deal with stationary subsets of [
2]
0
instead of
1
.
1.3 Corollary. If B is an
1
free Borel abelian group, then B is
2
free if and only
if K B : [K[ =
0
and B/K is
1
free is not stationary.
1.4 Fact: If 2
0
< 2
1
then every Borel Whitehead group B is
2
free.
Proof. By [DvSh 65] (or see [EM]) as 2
0
< 2
1
we have: if G be a Whitehead group
of cardinality
1
(hence is
1
free) and G =
_
<
1
G
is such that G
: <
1
) is
an increasing continuous chain of countable subgroups, then : G
+1
/G
is not
free does not contain a closed unbounded set (see [EM, Ch.XII,1.8]). Thus, if B
is not
2
free, then the subgroup G constructed in the proof of lemma 1.2 is not
Whitehead. Since being Whitehead is a hereditary property (see [EM]), B is not
Whitehead.
1.4
The lemma shows that
1.5 Conclusion. For Borel abelian groups B
, B
is
2
free is absolute (in fact
it is a
1
1
property of
).
Proof. The formula will just say that there is a model of a suitable fragment of
ZFC (e.g. ZC) with standard to which
belongs and it satises B
is
2
free.
1.5
(
4
0
2
)
r
e
v
i
s
i
o
n
:
1
9
9
9

0
8

1
2
m
o
d
i
f
i
e
d
:
2
0
0
3

0
3

0
7
BOREL WHITEHEAD GROUPS 7
2 On
2
free Whitehead
2.1 Theorem. If B is a Borel Whitehead group, then B is
2
free.
2.2 Conclusion: (CH or just 2
0
< 2
1
) Every Whitehead Borel abelian group is
free.
Before we prove we quote [Sh 44, Denition 3.1].
2.3 Denition. 1) If L is a subset of the
1
free abelian group, G, PC(L, G) is the
smallest pure subgroup of G which contains L. Note that if H is a pure subgroup
of G, L H then PC(L, G) = PC(L, H). We omit G if it is clear.
2) If H is a subgroup of G, L a nite subset of G, a G, then the statement
(a, L, H, G) means that: PC(H L) = PC(H) PC(L) but for no b PC(H
L a) do we have PC(H L a) = PC(H) PC(L b).
Proof. Assume B is not
2
free. We repeat the proof of Lemma 1.2. So in V
P
, B is
a nonfree
1
free abelian group of cardinality
1
. Hence by [Sh 44, p.250,3.1(3)],
B satises possibility I or possibility II where we have chosen
B = B
: <
1
) increasing continuous with B
; the
possibilities are explained below. The proof splits into the two cases.
Possibility I: By [Sh 44, p.250].
So we can nd (still in V
P
) an ordinal <
1
and a
i
B for i <
1
, < n
i
such
that
(A) a
i
+B
: i <
1
, n
i
is independent in B/B
(B) (a
n
i
, L
i
, B
, B) where L
i
= a
i
: < n
i
.
This situation does not survive well under the process and the proof of Lemma 1.2
but after some analysis a revised version will.
Without loss of generality n
i
= n() = n
, B, B
: <
1
),
a
i
0
, . . . , a
i
n
i
) : i <
1
_
belong to N. We can nd M V, M
= N; without loss of generality M is transi
tive (so M [= n is a natural number i n is a natural number). We now work in
V .
Let B (H (), , <
M
). We can nd t
(v) :
2) such that:
(a) each t
0
(v), . . . , t
k1
(v)) is generic too (for
k
), i.e. if D B is a dense subset
of
k
then
<k
t
, M M
, M
such that y
realizes t
(v) in M
. So M
[= y
a
countable ordinal. Without loss of generality if M
[=
2 then
2 and
(n) = i M
! As a =
a
i
: n
) : i <
1
_
N we can look at a and h( a) as a two
place function (with variables written as superscript and subscript). So we can let
a
( n
,
2) be reals such that: M
[= h( a)
y
= a
. By absoluteness
a
B (more exactly a
= B
, a
n
/E
B) and (a
n
, a
: < n
), B
, B).
If we can prove that a
:
2, n
) is independent over B
(= h(B
)), then
the proof of [Sh:98, 3.3] nish our case: proving B is not Whitehead group. But
independence is just a demand on every nite subset. So it is enough to prove
if k < ,
0
, . . . ,
k1
2 are distinct, then
a
: n
.
We prove this by induction on k. For k = 0 this is vacuous, for k = 1 it is part of
the properties of each a
: n
0
(v), . . . , t
k
(v)) (more exactly
k
t
(v)
t
(v) and s
m
Z for m k, n
such that:
if t
m
(v) is generic over B for m k, moreover t
m
(v) : m k) is a
generic subset of
k+1
over B and
m
(v) t
m
(v), then (dening M
m
by
t
m
(v) and a
as before)
mk
s
m
= t B
.
Clearly for m k we have M [= v :
m
(v)v a countable ordinal has order type
1
and without loss of generality also M [= v : M [=
m
(v) v a countable
ordinal has order type
1
.
So in M there are g
0
, . . . , g
k
M such that: M [= g
i
is a permutation of
1
,
for i k we have (v)(
0
(v)
0
(g
i
(v)) and g
0
(v), g
1
(v), . . . , g
k
(v) are pairwise
distinct. Let for m k, t
i
0
(v) = (v) : (g
i
(v)) t
0
(v). Let in M
0
, y
i
0
=
[g
i
(y
0
)]
M
0
, a
0
,i
= [h( a)
(y
i
0
)
]
M
0
. Now y
i
0
realizes t
i
0
(v) and M
0
is also the
Skolem hull of M y
i
0
and t
i
0
(v), t
1
(v), . . . , t
k
(v))
k+1
is generic over B
and
0
(v) t
i
0
(v),
1
(v) t
1
(v), . . . ,
k
(v) t
k
(v). Hence for each i k in B
we have
s
0
0
,i
0<mk
n
s
m
= t B
.
By linear algebra a
0
,i
: i k, n
0
. But the
(
4
0
2
)
r
e
v
i
s
i
o
n
:
1
9
9
9

0
8

1
2
m
o
d
i
f
i
e
d
:
2
0
0
3

0
3

0
7
BOREL WHITEHEAD GROUPS 9
formula saying this is false holds in (H (), , <
(it speaks on a, B, B
for n
such that (a
, a
: <
n
, B
, B) is stationary; all in V
P
. Now without loss of generality we can nd
n
: n < ) such that:
n
<
n+1
, =
_
n<
n
, and there are y
m
B
+1
, t
m
B
n
+1
and s
m,
Z, (for < n
) such that:
()
0
y
0
= a
and
()
2
s
m,n
m+1
=
<n
m,
a
+y
m
+t
m
()
3
s
m,n
m,n
(e.g. 1) then sy
m+1,n
is not in B
+a
i
: < n
m
)
B
()
4
if
n
: n < then PC
B
(B
+1
a
0
, . . . , a
) =
PC
B
(B
0
, . . . , a
) +B
+1
[why? known, or see later.]
Without loss of generality S n
= n
) such that a =
: n
) : S
_
, =
n
: n < ) : S
_
,
_
(s
m,
: n
), y
m
, t
m
)
m<
: S
_
belong to N, then de
ne M and choose B as before. We let this time =
M
be as in the proof of
Lemma 1.2, that is (v) : M [= S : () stationary. Now we work in V .
We can nd t
(v) :
2) such that:
(a) each t
0
, . . . , t
k1
) is generic
for
k
over B
(c) letting M
, y
be such that: M M
, M
, y
realizes t
(v) in M
we have
(i) M
[= y
is a countable ordinal S
(ii) M [= a is a countable ordinal M
[= a < y
(iii) if y M
[= y
y.
So looking at h : N M the isomorphism, then
n
=: [h( )]
y
n
for n < satises:
M
[=
n
a countable ordinal
M
[=
n
<
n+1
< y
[= the set[h( )]
y
n
: n < is unbounded below y
(
4
0
2
)
r
e
v
i
s
i
o
n
:
1
9
9
9

0
8

1
2
m
o
d
i
f
i
e
d
:
2
0
0
3

0
3

0
7
10 SAHARON SHELAH
hence
n
: n < M is unbounded among the countable ordinals of M.
Now by easy manipulation (see proof below):
(c) if
1
,=
2
2 then
1
n
: n <
2
n
: n < is nite.
(We can be lazy here demanding just that no
n
: n < is included in the union
of a nite set with the union of nitely many sets of the form
n
: n < where
2, which follows from pairwise generic, and one has to do slightly more
abelian group theory work below).
Now we can let a
= [(h( a))
y
]
M
:
2 and n
: a B
M hence a
contradiction to our being in possibility II (or directly get in the proof in the
case possibility I holds).
An alternative is the following:
We are assuming that in V
P
, possibility I fails. So also in V , letting A = M B
: n) : n < , a
B, a
:
n) independent over A in B and (a
n
, a
n
s
) :B [= (y)(sy =
),
s, s
integers
_
so t( a, A) : a K[A] is countable. But for the
2 the types
t(a
: < n
M,n
=
_
(v) :(i) (v) is a rst order formula with parameters from M
(ii) for some
M
1
for < n we have
M [= (v)((v) v h(S)) &
<n
(h( ))
v
)
(iii) M [= ( <
1
)(
stat
v <
1
)[((v) & < (h( ))
v
n
)]
_
.
Now note:
0
M,n
M
1
if (v)
M
and n < then for some m [n, ) and
M
1
for
< m we have (v) &
<m
(h( ))
v
belongs to
M,m
2
if (v)
M,n
and M
1
then
(v) :
n
2), k
()
(v) D
if < g()
()
(v)
i
(v) for i = 0, 1
() if
0
,=
1
n
2,
i
i
n+1
2 for i = 0, 1 and k
0
k < k
0
and M [=
(v)(
0
(v) (h( ))
v
k
= ) then M [= (v)[
1
(v)
<k
1
(h( ))
v
,= ].
There is no problem to do it and t
(v) = (v)
M
: (v)
M
n
(v) for some
n < for
2 are as required.
Why does hold?
For S let w
= < : PC
B
(B
+1
a
0
, . . . , a
) is not equal to PC
B
(B
0
, . . . , a
n,
) +B
+1
B.
Let S
= S : ( < )([w
[ <
0
), if S
is stationary we get ,
otherwise SS
let
= Min : w
is innite.
By Fodors lemma for some () <
1
, S
= SS
= () is stationary
hence uncountable and we can get possibility I, contradiction.
2.1
(
4
0
2
)
r
e
v
i
s
i
o
n
:
1
9
9
9

0
8

1
2
m
o
d
i
f
i
e
d
:
2
0
0
3

0
3

0
7
12 SAHARON SHELAH
3 Refinements
We may wonder if we can weaken the demand Borel.
3.1 Denition. 1) We say
is a code for a Souslin abelian group if in Denition
0.1 we weaken the demand on
0
,
1
to being a
1
1
relation.
2) A model M of a fragment of ZFC is essentially transitive if:
(a) if M [= x is an ordinal and (y : y <
M
x,
M
) is well ordered then x is
an ordinal and M [= y x y x
(b) if is an ordinal, (y : y <
M
,
M
) is well ordered and M [= an
ordinal, rk(x) = , then M [= y x y x.
3) For M essentially transitive with standard such that
M let B
M
is B
as
interpreted in M and trans(M) = x M : x as in (b) of part (2).
3.2 Fact. 1)
)
M
trans(M) there is a homomorphism j
M
from B
M
into B = B
implies j
M
(t) = x/E
.
4) If M N are as in (3), then j
M
j
N
.
Proof. Straightforward.
3.3 Claim. 1) In 1.2, 2.1 we can assume that B = B
is only Souslin.
2) If B = B
is not
2
free, then case I of [Sh 44](3.1) holds, moreover the con
clusion of case I in the proof of 2.1 holds.
Remark. If only
1
is Souslin, i.e. is
1
1
, just repeat the proofs.
Proof. For both we imitate the proof of 2.1.
In both possibilities, for each
2, let G
,
(the M
s chosen as there). So j
M
is a homomorphism from G
into B. However,
j
M
j
M
and j
M
is one to one (noting that h, the unique isomorphism from N
onto M, is the identity on (
2) N, hence on B
N, and also B
V
= B
V
P
). Let
B
= Rang(j
M
). Now in dening (x, L, B
PC(B
L x) such that (x
, L
, B
, B) and [L
a
n
: n
, B)
M
is
one to one and by easy algebraic argument, we can get, for 2.1, nonWhiteheadness
and for 1.2, non
2
freeness.
3.3
(
4
0
2
)
r
e
v
i
s
i
o
n
:
1
9
9
9

0
8

1
2
m
o
d
i
f
i
e
d
:
2
0
0
3

0
3

0
7
BOREL WHITEHEAD GROUPS 13
3.4 Fact. 1) B
is non
2
free is a
1
1
property of
, assuming B
is a
1
free
Souslin abelian group.
2)
codes a
1
free Souslin abelian group is a
1
2
property of
.
Proof. Just check.
(
4
0
2
)
r
e
v
i
s
i
o
n
:
1
9
9
9

0
8

1
2
m
o
d
i
f
i
e
d
:
2
0
0
3

0
3

0
7
14 SAHARON SHELAH
REFERENCES.
[BKM78] Jon Barwise, Matt Kaufmann, and Michael Makkai. Stationary logic.
Annals of Mathematical Logic, 13:171224, 1978.
[DvSh 65] Keith J. Devlin and Saharon Shelah. A weak version of which follows
from 2
0
< 2
1
. Israel Journal of Mathematics, 29:239247, 1978.
[EM] Paul C. Eklof and Alan Mekler. Almost free modules: Set theoretic meth
ods, volume 46 of NorthHolland Mathematical Library. NorthHolland
Publishing Co., Amsterdam, 1990.
[Sch85] J. Schmerl. Transfer theorems and their application to logics. In
J.Barwise and S.Feferman, editors, Model Theoretic Logics, pages 177
209. SpringerVerlag, 1985.
[Sh 44] Saharon Shelah. Innite abelian groups, Whitehead problem and some
constructions. Israel Journal of Mathematics, 18:243256, 1974.
[Sh 43] Saharon Shelah. Generalized quantiers and compact logic. Transactions
of the American Mathematical Society, 204:342364, 1975.
[Sh:98] Saharon Shelah. Whitehead groups may not be free, even assuming CH.
II. Israel Journal of Mathematics, 35:257285, 1980.
[Sh 202] Saharon Shelah. On coSouslin relations. Israel Journal of Mathemat
ics, 47:139153, 1984.
[Sh 161] Saharon Shelah. Incompactness in regular cardinals. Notre Dame Jour
nal of Formal Logic, 26:195228, 1985.
[Sh 523] Saharon Shelah. Existence of Almost Free Abelian groups and re
ection of stationary set. Mathematica Japonica, 45:114, 1997.
math.LO/9606229.
[Sh:f] Saharon Shelah. Proper and improper forcing. Perspectives in Mathe
matical Logic. Springer, 1998.