7

5
4


r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
0
-
3
0







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


IT IS CONSISTENT WITH ZFC THAT B
1
-GROUPS ARE NOT B
2
SAHARON SHELAH AND LUTZ STR

UNGMANN
Abstract. Both, B
1
-groups and B
2
-groups are natural generalizations of nite rank
Butler groups to the innite rank case and it is known that every B
2
-group is a B
1
-
group. Moreover, assuming V = L it was proven that the two classes coincide. Here we
demonstrate that it is undecidable in ZFC whether or not all B
1
-groups are B
2
-groups.
Using Cohen forcing we prove that there is a model of ZFC in which there exists a
B
1
-group that is not a B
2
-group.
1. Introduction
The study of Butler groups, both in the nite and in the innite rank case, is a
most active area of Abelian Group Theory. There are several challenging problems
which require deep insight into the theory of Butler groups and the available methods
as well as the development of new machinery. The nite rank case is closely related
to the study of representations of nite posets while the innite rank case has its own
special avor. During the last years more and more the connection between innite rank
Butler groups and innite combinatorics was discovered and led to numerous interesting
results. In this paper we discuss one of the long-standing problems, namely whether or
not all B
1
-groups are B
2
-groups, and show that its solution is independent of ZFC. It
is known that any B
2
-group is a B
1
-group and moreover, assuming Goedels universe of
constructibility the two classes coincide. In contrast to this result we will show, using
Cohen forcing, that there is a model of ZFC in which there exists a B
1
-group that is
not a B
2
-group.
In the following all groups are abelian. Our terminology is standard and maps are
written on the left. If H is a subgroup of a torsion-free group G then the purication
of H in G is denoted by H

. For notations and basic facts we refer to [11] for abelian

groups, [18] and [21] for forcing and [9] or [17] for set-theory. Moreover, the interested
reader may look at [2] for a survey on nite rank Butler groups and at [3], [12] for
surveys on innite rank Butler-groups.
Since our problem comes from abelian group theory the authors tried to make the paper
1991 Mathematics Subject Classication. 20K20, 05E99.
Key words and phrases: B
1
-groups, B
2
-groups, consistency result, Cohen forcing
The rst author was supported by project No. G-0545-173,06/97 of the German-Israeli Foundation for
Scientic Research & Development.
Publication 754 in Shelahs list of publications
The second author was supported by a MINERVA fellowship.
1
7
5
4


r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
0
-
3
0







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


2 SAHARON SHELAH AND LUTZ STR

UNGMANN
accessible for non set-theorists. Hence the involved set-theory (forcing) is explained in
detail although the methods are very standard.
2. Infinite rank Butler groups
In this section we recall the denitions of B
1
-groups and B
2
-groups as they were given
by Bican-Salce in [6]. Both classes contain the class of nite rank Butler-groups (pure
subgroups of completely decomposable groups of nite rank) rst studied by Butler in
[4]. Let us begin with the notion of a balanced subgroup.
A pure subgroup Aof the torsion-free group Gis said to be a balanced subgroup if every
coset g +A (g G) contains an element g +a (a A) such that (g +a) (g +x) for
all x A, where (g) denotes the characteristic of an element g G. Such an element
is called proper with respect to A and (g) denotes the characteristic of an element g in
the given group G.
An exact sequence 0 A G C 0 is balanced exact if the image of A in
G is a balanced subgroup of G. Hunter [16] discovered that the equivalence classes of
balanced extensions of a group H by a group G give rise to a subfunctor Bext
1
(H, G)
of Ext
1
(H, G) and hence homological algebra is applicable. Thus for a balanced exact
sequence
() 0 A G C 0
and a group H we obtain the two long exact sequences
0 Hom(C, H) Hom(G, H) Hom(A, H) Bext
1
(C, H) Bext
1
(G, H)
Bext
1
(A, H) Bext
2
(C, H)
and
0 Hom(H, A) Hom(H, G) Hom(H, C) Bext
1
(H, A) Bext
1
(H, G)
Bext
1
(H, C) Bext
2
(H, A)
It is routine to check that balanced-exactness of the sequence () is equivalent to the
following property: for every rank one torsion-free group R, every homomorphism R
C can be lifted to a map R G, i.e. every rank one torsion-free group is projective
with respect to (). Thus the following lemma is easily established.
Lemma 2.1. Let
0 A G

C 0
be a balanced exact sequence. Then this sequence is locally invertible, i.e. for any element
c C there exists a homomorphism
c
: c)

G such that
c
= id
c

.
We now come to the denitions of B
1
-groups and B
2
-groups.
Denition 2.2. A torsion-free abelian group B is called
(1) a B
1
-group if Bext
1
(B, T) = 0 for all torsion groups T;
7
5
4


r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
0
-
3
0







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


IT IS CONSISTENT WITH ZFC THAT B1-GROUPS ARE NOT B2 3
(2) a B
2
-group if there exists a continuous well-ordered ascending chain of pure sub-
groups,
0 = B
0
B
1
B

= B =

<
B

such that B
+1
= B

+ G

for every < for some nite rank Butler group

G

; i.e. B

is decent in B
+1
in the sense of Albrecht-Hill [1];
(3) nitely Butler if every nite rank pure subgroup of B is a Butler-group.
Due to Bican-Salce [6] the three denitions are equivalent for countable torsion-free
groups.
Theorem 2.3 ([6]). For a countable torsion-free abelian group B the following are
equivalent:
(1) B is nitely Butler;
(2) B is a B
2
-group;
(3) B is a B
1
-group.
Without any restriction to the cardinality we have in general:
Theorem 2.4 ([6]). B
2
-groups of any rank are B
1
-groups.
It turned out that the converse implication in the above theorem couldnt be proved
without any additional set-theoretic assumptions. There are some partial results in
ZFC characterizing the B
2
-groups among the B
1
-groups but none of them is really
satisfactory. The following was shown by Fuchs and Rangaswamy independently.
Lemma 2.5 ([13], [20]). Suppose that 0 H C G 0 is a balanced-exact
sequence where C is a B
2
-group and H and G are B
1
-groups. If one of H and G is a
B
2
-group, then so is the other.
An attempt to characterize the B
2
-groups in a homological way is the following the-
orem due to Fuchs.
Theorem 2.6 ([13]). If B is a B
2
-group, then Bext
i
(B, T) = 0 for all i 1 and for all
torsion groups T.
Assuming the continuum hypothesis Rangaswamy was able to show that also the
converse holds and in some cases Fuchs could even remove CH.
Theorem 2.7 ([13], [20]). The following are true:
(1) Assuming CH a torsion-free group B is a B
2
-group if and only if Bext
1
(B, T) =
Bext
2
(B, T) = 0 for all torsion groups T.
7
5
4


r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
0
-
3
0







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


4 SAHARON SHELAH AND LUTZ STR

UNGMANN
(2) A torsion-free group B of cardinality
n
(for some integer n 1) is a B
2
-group
if and only if Bext
i
(B, T) = 0 for all i n + 1 and all torsion groups T.
It was natural to ask whether Bext
2
(B, T) is always zero for a torsion-free group B
and a torsion group T but Magidor-Shelah [19] proved that this is not the case even
assuming the generalized continuum hypothesis GCH. That CH was relevant in many
papers was explained by Fuchs who showed the following theorem.
Theorem 2.8 ([13]). In any model of ZFC, the following are equivalent:
(1) Bext
2
(G, T) = 0 for all torsion-free groups G and torsion groups T;
(2) CH holds and balanced subgroups of completely decomposable groups are B
2
-
groups.
One of the most interesting and main results in the theory of innite rank Butler
groups is the following nal theorem of this section proved by Magidor and Fuchs.
Theorem 2.9 ([14]). Assuming V = L every B
1
-group is a B
2
-group.
We will show in this paper that the conclusion of the last theorem does not hold in
ZFC but is independent of ZFC.
3. The forcing
In this section we will explain the forcing notion we are going to use to construct our
B
1
group H which fails to be B
2
. The reader who is familiar with forcing, especially
with adding Cohen reals may skip this section. Most results are well-known and basic.
For unexplained notations and further results on forcing we refer to Kunens book [18]
or to the more advanced rst authors book [21].
Let M be any countable transitive model of ZFC and assume of course that the set
theory ZFC is consistent. The aim of forcing is to extend M to a new model which still
satises ZFC but which has additional properties which we are interested in.
A forcing notion P M is just a non empty, preordered set (P, , 0
P
), where 0
P
is the
minimal element of P, hence 0
P
p for all p P. Note that we dont require that p q
and q p imply q = p. If two elements p, q P have no common upper bound, i.e.
there is no t P such that p t and q t, then we say that p and q are incompatible
and write p q. If a common upper bound exists we call the elements compatible. We
now want to add to M a subset S of P to construct a transitive set M[S] which is a
model of ZFC with the same ordinals as M such that M M[S] and S M[S]. Those
sets S are called generic.
Denition 3.1. Let D P, S P and p P. Then
(1) D is called dense in P if for any q P there is an element t D such that q t;
(2) D is dense above p if for any q P such that p q there exists an element
t D such that q t;
7
5
4


r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
0
-
3
0







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


IT IS CONSISTENT WITH ZFC THAT B1-GROUPS ARE NOT B2 5
(3) S is called P-generic over M if the following hold:
(a) for all q, r S there exists t S such that q t and r t, i.e. all elements
of S are compatible in S;
(b) if q S and t q for some t P then also t S;
(c) S D ,= for every dense subset D of P which is in M.
A rst observation is that a generic set S intersects non-trivially also sets which are
dense above p in many cases.
Lemma 3.2. Let D P and S be P-generic over M. Then
(1) Either S D ,= or there exists q S such that for all r D we have r q;
(2) If p S and D is dense above p, then S D ,= .
Proof. See [18, Lemma 2.20].
If S is P-generic over M (or, for short, generic), then the existence of the model M[S]
with the desired properties follows from the Forcing Theorem (see [21]). M[S] is the
smallest transitive model of ZFC that contains M and S. We dont want to recall the
construction of M[S] but we would like to mention the following facts. Since we want
to prove theorems in M[S] we would like to know the members of M[S] but we can
not have full knowledge of them inside M since this would cause these sets to be in M
already. If S is in M then M[S] gives nothing new, so we have to assume that S is not
in M and this is the case in general as the following lemma shows.
Lemma 3.3. Let S be P-generic over M. If P satises the following condition
(3.1) p P q, r P such that p q, p r and q r
then S , M.
Proof. See [18, Lemma 2.4].
Nevertheless, every element p of P can be a member of a generic set.
Lemma 3.4. Let p P. Then there is a subset S which is P-generic over M such that
p S.
Proof. See [18, Lemma 2.3].
Although we dont know the generic set S we assume that we have some prescription
for building the members of M[S] out of M and S. These prescriptions are called P-
names, usually denoted by , and their interpretation in M[S] is [S]. For the exact
7
5
4


r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
0
-
3
0







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


6 SAHARON SHELAH AND LUTZ STR

UNGMANN
denition of P-names and their interpretation we refer again to Kunens book [18] but
let us mention that the Strengthened Forcing Theorem (see [21]) shows that
M[S] = [S] : M and is a P-name .
If we are talking about the P-name of a special object H from M[S] without specifying
S then we will write

H instead of H to avoid confusion but if H is already in M, then
we omit the tilde. Any sentence of our forcing language uses the P-names to assert
something about M[S] but the truth or falsity of a sentence in M[S] depends on S
in general. If p P, then we write p and say p forces to mean that for all S
which are P-generic over M, if p S, then is true in M[S]. If 0
P
then we just
write
P
which means that for any generic S the sentence is true in M[S] since 0
P
is always contained in S. Hence the elements of P provide partial information about
objects in M[S] but not all information and if p q then q contains more information
than p. It may be decided in M whether or not p and whenever a sentence is
true in M[S] then there is p S such that p .
We now turn to the forcing of adding Cohen reals. Therefore let be an uncountable
cardinal. We put
P = p [ p is a function from a nite subset of to 2
= p [ p : dom(p) 2, dom(p) a nite subset of
The partial ordering of P is given by set theoretic inclusion, i.e. two functions p and
q satisfy p q if and only if q extends p as a function. This forcing is called adding
Cohen reals and the elements of P can obviously be regarded as functions from to
<
2 which we will do in the sequel.
The next lemma shows why the forcing is called adding Cohen reals.
Lemma 3.5.
P
There are at least reals.
Proof. See [21, Chapter I, Lemma 3.3].
We will give the Cohen reals P-names, say

for < and state some basic

properties of the Cohen reals. Note that a real is a function from to 2 = 0, 1.
Lemma 3.6. The following hold for , < :
(1)
P
There are innitely many n N such that

(n) =

(n) = 1;
(2)
P
There are innitely many n N such that

(n) =

(n) = 0;
(3)
P
There are innitely many n N such that

(n) ,=

(n).
7
5
4


r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
0
-
3
0







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


IT IS CONSISTENT WITH ZFC THAT B1-GROUPS ARE NOT B2 7
Proof. The proof of this fact is standard using a densitiy argument.
Moreover, we have three more important facts.
Lemma 3.7. The following hold for P.
(1) P satises the c.c.c. condition, i.e. P has no uncountable subset of pairwise
incompatible members;
(2) P preserves cardinals and conalities, i.e. if is a cardinal in M, then is also
a cardinal in M[S] with the same conality;
(3)
P
2

0
. In particular, if

0
= in M, then
P
2

0
= .
Proof. See [21, Chapter I, Lemma 3.8], [21, Chapter I, Theorem 4.1] and [18, Theorem
5.10].
Finally we would like to remark that our notation is the Jerusalem style of forcing
notation like in [21] but diers from the notation for example in [18]. In our partial
order p q means that q contains more information than p does and not vice versa.
4. Our B
1
group H
Let M be a countable transitive model of ZFC in which the generalized continuum
hypothesis holds, i.e. 2

=
+
for all innite cardinals . Moreover, let
4
be
regular and let P be the forcing of adding Cohen reals. As we have seen in the last
section, P preserves cardinals and conalities and 2

0
= in M[S] for every generic S.
Let

denote the Cohen reals for < and let M

be a model extending M, obtained

by Cohen forcing, e.g. M

= M[S] for some xed S.

Inside M

we choose independent elements

x
n
: n < and y

: <
and x a countable set of natural prime numbers
p
n
: n <
such that p
n
< p
m
for n < m.
Denition 4.1. Let W =

n<
Qx
n


<
Qy

be the rational vector space and let F =

n<
Zx
n


<
Zy

be the free abelian group generated by the x

n
s and y

s. In M

we
dene H as the subgroup of W generated by F p
1
n
(y

x
n
) : < , n < ,

(n) = 1
and let

H be its P-name.
We can now state our Main Theorem.
7
5
4


r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
0
-
3
0







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


8 SAHARON SHELAH AND LUTZ STR

UNGMANN
Main Theorem 4.2. In the model M

the group H is a B
1
-group but not a B
2
-group.
Hence it is consistent with ZFC that B
1
-groups need not be B
2
-groups.
The proof of the Main Theorem 4.2 will be divided into two parts. The rst part is
to show that H is a B
1
-group which will be done in this section. Section 5 will then
consist of proving that H is not B
2
.
Theorem 4.3. In the model M

the group H is a B
1
-group.
The proof of Theorem 4.3 takes the rest of this section and consists of several steps.
Proof. (of Theorem 4.3) To prove that H is a B
1
-group we have to show that Bext(H, T) =
0 for any torsion group T. Hence let
(4.2) 0

T
id

H 0
be forced to be a balanced exact sequence in M

with T =

T[S] torsion. Thus there
exists r

P such that
r

0

T
id

H 0 is balanced exact.
We now work in M

and choose preimages g

G of y

under for all < . Similarly

let x
n
G be a preimage for x
n
under for n < . Moreover, let
A

= n < :

(n) = 1
for < .
It is our aim to show that the balanced exact sequence (4.2) is forced to split, hence
it is enough to prove that the homomorphism is right-invertible, i.e. we have to nd
: H G such that = id
H
. Therefore it is necessary to nd preimages of the
generators of H in G such that equations satised in H also hold in G. We need the
following denition.
Denition 4.4. Let < and t T arbitrary. Then the set R
,t
is dened as
R
,t
= n A

: g

t x
n
is not divisible by p
n
.
We will now use a purely group theoretic argument to show that if for every <
there is a t

T such that R
,t
is nite then is invertible.
Lemma 4.5. Let < and let t T such that R
,t
is nite. Then there exists t

T
such that R
,t
= .
7
5
4


r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
0
-
3
0







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


IT IS CONSISTENT WITH ZFC THAT B1-GROUPS ARE NOT B2 9
Proof. Since R
,t
is nite we may assume without loss of generality that R
,t
has min-
imal cardinality. Assume that R
,t
is not empty and x n R
,t
. By the primary
decomposition theorem we decompose T as
T = T
pn
T

where T
pn
denotes the p
n
-primary component of T. Since n A

it follows that p
n
divides y

x
n
, hence there exists z G such that
(z) = p
1
n
(y

x
n
).
Thus
(g

t x
n
) p
n
z T = T
pn
T

and therefore there exist t

0
T
pn
and t
1
T

such that
(g

t x
n
) p
n
z = t
0
+t
1
.
Since T

is divisible by p
n
we can write t
1
= p
n
t
2
for some t
2
T

. Hence
(g

t x
n
) p
n
(z t
2
) = t
0
.
We let t

= t + t
0
and will show that R
,t
has smaller cardinality than R
,t
- a contra-
diction. By the choice of t

we have
(g

x
n
) = g

t t
0
x
n
= p
n
(z t
2
)
and hence n , R
,t
. But on the other side, if m , R
,t
, then p
m
divides (g

t x
m
)
and thus p
m
divides (g

(t

t
0
) x
m
). Since p
n
,= p
m
it follows that p
m
divides t
0
and therefore p
m
divides (g

x
m
). Hence m , R
,t
showing that R
,t
is strictly
smaller than R
,t
. This nishes the proof.
Lemma 4.6. Assume that for every < there exists t

T such that R
,t
is nite.
Then is invertible and hence the sequence (4.2) is forced to split.
Proof. By Lemma 4.5 we may assume without loss of generality that for every <
the set R
,t
is empty. Thus for each n A

we can nd z
,n
G such that
p
n
z
,n
= g

x
n
t

.
We now dene a homomorphism : H G as follows:
(1) (x
n
) = x
n
(n < );
(2) (y

) = g

( < );
(3) (p
1
n
(y

x
n
)) = z
,n
( < , n A

).
7
5
4


r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
0
-
3
0







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


10 SAHARON SHELAH AND LUTZ STR

UNGMANN
We leave to the reader to check that (1), (2) and (3) induce a well-dened homomorphism
: H G satisfying = id
H
.
(Continuation of the proof of Theorem 4.3) Up to now we havent used any forcing
but we have worked in the model M

. By Lemma 4.6 it remains to nd for every <

an element t

T such that the set R

,t
is nite. Here we use the forcing.
We dene for ,= < the pure subgroup H
,
= y

of H. Since the
sequence (4.2) is forced to be balanced exact Lemma 2.1 shows that there exist homo-
morphisms

,
: H
,
G such that
,
= id
H
,
.
Let h
,
=
,
(y

) G, hence
t
,
= h
,
(g

) T.
Since T is a torsion group we can nd m
,
< such that
ord(t
,
) = m
,
.
Let m
,
and g

, g

be P-names for m
,
and g

, g

, respectively. We can now easily

show
Fact 4.7. r

If n > m
,
, then p
n
divides ( g

) for n A

Proof. If n > m
,
, then p
n
> m
,
follows since the primes p
m
are increasing. Therefore
gcd(p
n
, m
,
) = 1 and thus p
n
divides (h
,
(g

)). Moreover, h
,
=
,
(y

)
is divisible by p
n
since n A

. Hence p
n
divides (g

).
Now let r

r
,
P be such that r
,
forces the value m
,
to m
,
, i.e.
r
,
m
,
= m
,
.
Without loss of generality we assume that dom(r
,
) for all , . Since all elements
of P are functions from to 2 with nite domain, we may write for some n
,
<
dom(r
,
) =
(,,0)
, ,
(,,n
,
)
,
where
(,,i)
<
(,,j)
if i < j n
,
. We would like to apply the -Lemma to the
functions r
,
to obtain a -system but unfortunately the functions r
,
depend on two
variables. This forces us to do the -Lemma by hand. For this we use the Erd os-Rado
Theorem (see [10]).
First we dene a coloring on 4-tuples in
4
. Let
0
,
1
,
2
,
3
<
1
such that
0
<

1
<
2
<
3
and let
c(
0
,
1
,
2
,
3
)
7
5
4


r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
0
-
3
0







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


IT IS CONSISTENT WITH ZFC THAT B1-GROUPS ARE NOT B2 11
consist of the following entries in an arbitrary but xed order:
(i ) n

0
,
1
;
(ii ) m

0
,
1
;
(iii ) r

0
,
1
(
(
0
,
1
,j)
: j n

0
,
1
);
(iv ) (tv(
(n
1
,n
2
,n
3
)
<
(m
1
,m
2
,m
3
)
) : n
1
, n
2
, m
1
, m
2
< 4; n
3
< n
n
1
,n
2
; m
3
< n
m
1
,m
2
).
Recall that tv denotes the truth-value of the inequality. The above coloring is a coloring
with colors and thus we may apply the Erd os-Rado Theorem. Note that we are
working in our model M in which GCH holds by assumption. Hence we have

4
(
1
)
4

0
which is exactly what we need to apply the Erd os-Rado Theorem. We obtain an in-
creasing chain of c-homogeneous elements
=

: <
1

which means that whenever

1
,

2
,

3
,

4
such that

1
<

2
<

3
<

4
, then
c(

1
,

2
,

3
,

4
) = c

for a xed color c

. Let this particular color consist of the following entries:

(I ) n

;
(II ) m

;
(III ) (k
1
, , k
n
) (k
i
0, 1);
(IV ) (l
1
, , l
16
2
(n

)
2) (l
i
Yes, No).
Let us rst explain what the homogenity implies. Let

1
,

2
such that
1
<
2
,
then (I) ensures that the domain of r

1
,
2
has size n

. Moreover, (II) says that r

1
,
2
forces the value m

to m

1
,
2
and (III) implies that the image of r

1
,
2
is uniquely
determined. Finally (IV) ensures that if we take another pair

3
,

4
such that

3
<
4
, then the relationship between the elements of the domains of r

1
,
2
and r

3
,
4
is xed.
In the sequel we need to be above all the trouble, hence we may increase m

without
loss of generality such that m

is greater than or equal to length(r

,
(
,,e
)) for all
< <
1
and e n

. We can now approach to the -Lemma.

Denition 4.8. For

we dene
(1 ) u

= dom(r
,
+1
) dom(r
,
+2
);
(2 ) u

<
1
u

;
(3 ) s

= r
,
+1

u
= r
,
+2

u
.
7
5
4


r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
0
-
3
0







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


12 SAHARON SHELAH AND LUTZ STR

UNGMANN
We have to explain why (3) in Denition 4.8 is well-dened. This follows from ho-
mogenity since (IV) implies that for u

we have r
,
+1
() = r
,
+2
(). We are
now ready to show the following lemma, our version of the -system. Note that if we
talk about a -system of functions then we mean that the corresponding domains of
the functions form a -system.
Lemma 4.9. For

such that < we have

u

= u

.
Hence the functions s

) form a -system with root u

. Moreover, for xed

<
1
the functions r
,
( < <
1
) form a -system with root u

.
Proof. Let

be such that < . Clearly we have u

by Denition
4.8. It remains to show the converse inclusion. Therefore let u

and choose
<
1
arbitrary. We have to prove that lies in u

.
If = or = , then we are done.
If +1, then c(

,
+1
,

,
+1
) = c

by homogenity. Since dom(r

,
+1
) we
can nd i n

such that =
(,
+1
,i)
and similarly =
(,
+1
,j)
for some j n

.
It follows now that
tv(
(,
+1
,i)
<
(,
+1
,j)
) = No and tv(
(,
+1
,j)
<
(,
+1
,i)
) = No.
Hence there exists by homogenity k n

such that
tv(
(,
+1
,i)
<
( ,
+1
,k)
) = No and tv(
( ,
+1
,k)
<
(,
+1
,i)
) = No.
Thus =
( ,
+1
,k)
dom(r
,
+1
). Similarly it follows that dom(r
,
+2
) and
hence u

.
If < + 1, then we use similar arguments to those above to prove that u

.
Thus we have shown that u

for any <

1
and therefore u

.
The same kind of arguments show that also the functions r
,
( < <
1
) form a
-system with root u

for xed <

1
.
It is now easy to see by a pigeon-hole argument that we may assume without loss
of generality (and we will assume this in the sequel) that all the functions from the
-systems in Lemma 4.9 coincide on their root.
(Continuation of the proof of Theorem 4.3) The following denition now makes sense.
Denition 4.10. For < <
1
and a generic S P we dene
7
5
4


r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
0
-
3
0







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


IT IS CONSISTENT WITH ZFC THAT B1-GROUPS ARE NOT B2 13
(I ) s

= s

u
= s

(u
u
)
;
(II )

Y =

: s

S.
We can now show that s

is strong enough to force that

Y has cardinality
1
.
Fact 4.11. s

[

Y [=
1

Proof. Let

Y

= s

Y and assume that s

does not force

Y to be of size
1
.
Then [

Y

[=
1
. We will show that this set is dense above s

. Therefore let f s

,
then dom(f) is a nite subset of and u

dom(f). We choose s

such that
dom(s

)u

is disjoint to dom(f)u

. This is possible since by Lemma 4.9 the s

s form
a -system, hence
dom(s

)u

dom(s

)u

=
for ,= . Now, f and s

are compatible and thus

Y

is dense above s

. Therefore

S ,= by Lemma 3.2 - a contradiction.

We are almost done and prove the following statement.
Fact 4.12. s

If

Y and n A

[0, m

] then p
n
divides g

.
Proof. Let s

s be such that
s n A

[0, m

] .
Without loss of generality we may assume that s also forces truth values to

Y and

Y . If one of them is No, then we are done and hence let us assume that both are
Yes. We will show that there exists <
1
such that
(I ) > ;
(II ) > ;
(III ) dom(r
,
)u

dom(r
,
)u

is disjoint to dom(s).
Obviously we can choose > , such that dom(s) is disjoint to

, so all we have
to ensure is that also dom(r
,
)u

dom(r
,
)u

is disjoint to dom(s).
For this we prove that the three sets
(1 ) <
1
: dom(r
,
)u

is not disjoint to dom(s);

(2 ) <
1
: dom(r
,
)u

is not disjoint to dom(s);

(3 ) <
1
: u

is not disjoint to dom(s).

are bounded in
1
. Let us start with (1). By Lemma 4.9 we know that for each
<
1
the domains dom(r
,
) : < <
1
form a -system with root u

, hence
dom(r
,
)u

: < <
1
is a set of pairwise disjoint sets. Since dom(s) is a
7
5
4


r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
0
-
3
0







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


14 SAHARON SHELAH AND LUTZ STR

UNGMANN
nite set <
1
: dom(r
,
)u

is not disjoint to dom(s) must be bounded in

1
.
Similarly <
1
: dom(r
,
)u

is not disjoint to dom(s) is bounded in

1
. Finally,
again by Lemma 4.9 the sets u

: <
1
form a -system with root u

and so also
<
1
: u

is not disjoint to dom(s) is bounded in

1
.
For this we are able to prove that there is s
+
such that
(i ) s
+
s;
(ii ) s
+

(n) = 1;
(iii ) s
+
r
,
;
(iv ) s
+
r
,
.
Since n was chosen large enough which means that

has length less than or equal

to m

and hence less than or equal to n, there is, once we know that we can satisfy (i),
(iii) and (iv), also some s
+
s satisfying all conditions (i), (ii), (iii) and (iv). Thus
we only have to satisfy conditions (i), (iii) and (iv) and for this it is obviously enough
to show that the three functions s, r
,
and r
,
are compatible. Assume that r
,
and r
,
are incompatible, then by induction we obtain that r
,
1
and r
,
1
are
incompatible. Hence for < <
1
we have that r
,
1
and r
,
1
are incompatible
which contradicts the c.c.c. condition of our forcing. Therefore r
,
and r
,
are
compatible. Finally s and r
,
(and similarly s and r
,
) are compatible since by the
choice of we have dom(s) dom(r
,
) = u

.
Now s
+
p
n
divides g

and s
+
p
n
divides g

and therefore
s
+
p
n
divides g

as claimed.
Finally we have to prove another fact.
Fact 4.13. s

For every < there exists m

< such that m

< n A

and

Y such that n A

we have p
n
divides g

.
Let us rst show how Fact 4.13 implies Theorem 4.2, i.e. using Fact 4.13 we prove
that the set R
,0
is contained in [0, m

)Z for all < and hence nite after modifying

the choice of the preimages x
n
of x
n
(n < ) slightly (which doesnt has any eect on
what we have done so far). Choose x
n
G such that
(i ) ( x
n
) = x
n
;
(ii ) if n > m

and Y such that

(n) = 1, then p
n
divides g

x
n
.
For example if n > m

we choose Y such that

(n) = 1; Let k
n
G such that
(k
n
) = 1/p
n
(y

x
n
) and put x
n
= p
n
k
n
+g

. Then clearly (i) and (ii) are satised.

7
5
4


r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
0
-
3
0







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


IT IS CONSISTENT WITH ZFC THAT B1-GROUPS ARE NOT B2 15
Now Fact 4.13 ensures that R
,0
is contained in [0, m

) for if m

< n A

choose

Y such that

(n) = 1 (the one which was used when choosing the x

n
s), then we
have by the choice of x
n
that p
n
divides g

x
n
and by Fact 4.13 we have p
n
divides
g

and hence p
n
divides g

x
n
. Thus n , R
[,o)
and R
[,0)
[0, m

) follows.
Therefore the proof of Fact 4.13 nishes the proof of Theorem 4.3.
Proof. (of Fact 4.13) Fix < and let s
+
< s be such that s forces n A

. For
every <
1
we choose (if possible) t

in the generic set such that

(i ) s t

;
(ii ) t

Y ;
(iii ) t

m
,
= m

,
for some m

,
N.
Note that it is sucient to nd one such that
() p
n
divides g

g
,
for then we can use Fact 4.12 to get the conclusion for any

such that n A

. If we
have one t

satisfying (ii) and (iii), then it forces () for n > m

,
. So we rst ensure
(ii) and (iii) and then we use that there is an uncountable subset S

of
1
such that
t

: S

is a -system to ensure (i) where we put m

= m

,
which can be chosen
xed for the -system.
5. Why H fails to be B
2
To complete the proof of our Main Theorem 4.2 we show in this section that the group
H from Denition 4.1 can not be a B
2
-group in M

.
Theorem 5.1. In the model M

the group H can not be a B

2
-group.
Proof. Towards contradiction assume that H is a B
2
-group, hence has a B
2
-ltration
H =

<
H

.
Recall that a B
2
-ltration is a smooth ascending chain of pure subgroups H

such that
for every < H
+1
= H

+ B

for some nite rank Butler group B

. Recall that a
cub in is a subset C of such that
(i ) C is closed in , i.e. for all C

C, if sup C

< , then sup C

C;
(ii ) C is unbounded in , i.e. sup C

= .
The proof of the following lemma is standard (see [9, II.4.12]) but for the convenience
of the reader we include it briey.
7
5
4


r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
0
-
3
0







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


16 SAHARON SHELAH AND LUTZ STR

UNGMANN
Lemma 5.2. The set C = < [ H

= x
n
, y

: n < , < )

is a cub in .
Proof. First we show that C is closed in . Therefore let C

=
i
[ i I be a subset
of C such that sup C

< . If we put = sup C

, then clearly
H

iI
H

i
= x
n
, y

: n < , < )

and hence C.
It remains to show that C is unbounded. Therefore assume that C is bounded by

< , i.e.

for all C. We will show that there exists

< < such that

H

= x
n
, y

: n < , < )

, hence C - a contradiction.
Let
1
=

and put
E

1
= x
n
, y

: n < , <
1
)

.
Now choose
1

1
< such that E

1
H

1
. If
1
, C then choose
1

2
< such
that
H

1
E

2
= x
n
, y

: n < , <
2
)

.
Continuing this way we obtain a sequence of groups E

i
and H

i
such that
E

i
H

i
E

i+1
for all i < . Let = sup
i
: i < = sup
i
: i < . Then
H

i<
H

i
=

i<
E

i
= E

= x
n
, y

: n < , < )

and hence C. This nishes the proof.

(Continuation of the proof of Theorem 5.1) Now let C be such that >
1
and
w.l.o.g. let be a limit ordinal. This is possible since C is a cub by Lemma 5.2. Note
that y

, H

but we have the following lemma.

Lemma 5.3. There exists n

< and a sequence of ordinals

1

2

n
<
such that
H

+Zy

mn

B
m
+H

.
Proof. We induct on to show the even stronger statement that for any L

,
L of nite rank, there exist n

< and
1

2

n
< such that
H

+L)

mn

B
m
+H

.
If = , then we are done choosing n

= 1 and
1
= .
7
5
4


r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
0
-
3
0







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


IT IS CONSISTENT WITH ZFC THAT B1-GROUPS ARE NOT B2 17
If > is a limit ordinal, then L

implies L

for some < . Hence

we are done by induction hypothesis.
If = +1, let H

= H

+B

and let L = l
1
, , l
k
)

. We can nd representations
l
i
= h
,i
+b
,i
for all 1 i k where h
,i
H

and b
,i
B

. We put
L

= h
,i
, (B

) : 1 i k)

which is a pure subgroup of nite rank of H

. An easy calculation shows that L

L

+B

.
The induction hypothesis implies that there exist n < and
1

2

n
such that
H

+L

mn
B
m
+H

.
Another calculation shows that this implies
H

+L)

mn
B
m
+B

+H

.
This nishes the proof.
(Continuation of the proof of Theorem 5.1) By Lemma 5.3 we can choose n

< and

1

2

n
such that
H

+Zy

mn

B
m
+H

.
For every m n

we choose a nite set W

m
and an integer n
m
< such that
B
m

Wm
Zy

inm
Zx
i

.
Collecting all these generators and letting W =

mn

W
m
and k = maxn
m
: m n

we obtain
(5.3) H

+Zy

B +H

where B =

W
Zy

ik
Zx
i

.
Now choose < W and let n k be such that
n A

W,=
A

.
Note that this choice is possible by the following densitiy argument similar to the proof
of Lemma 3.6. Let p P and write p = (p

0
, p

1
, , p

m1
), where each p

i

<
2
7
5
4


r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
0
-
3
0







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


18 SAHARON SHELAH AND LUTZ STR

UNGMANN
(
i
< ). Without loss of generality we may assume that W
i
: i < m. Put
= max(W)+1 and let n = maxmax(dom(p

i
)) : i < m+1. Now we extend each p

i
to q

i
by putting q

i
(n) = 0. Moreover, we let q

(n) = 1. Then q = (q

0
, , q

m1
, q

)
is stronger than p and therefore forces what we need. It is now straightforward to
see that p
1
n
(y

) is an element of H

+Zy

but it is not an element of B + H

contradicting equation (5.3). This nishes the proof of Theorem 5.1 and therefore the
proof of our Main Theorem 4.2.
References
[1] U. Albrecht and P. Hill, Butler groups of innite rank and axiom 3, Czech. Math. J., 37, (1987),
293309.
[2] D.M. Arnold and C. Vinsonhaler, Finite rank Butler groups: a survey of recent results, Lecture
Notes in Pure Appl. Math., 146 (Marcel Dekker), (1993), 1741.
[3] L. Bican, Innite rank Butler groups, Advances in algebra and model theory, Gordon and Breach
Publishers, Algebra Log. Appl. 9 (1997), 287317.
[4] M.C.R. Butler, A class of torsion-free abelian groups of nite rank, Proc. London Math, Soc.
15 (1965), 680698.
[5] L. Bican and L. Fuchs, Subgroups of Butler groups, Comm. in Algebra, 22 , (1994), 10371047.
[6] L. Bican and L. Salce, Butler groups of innite rank, Abelian Group Theory, Lecture Notes in
Math. 1006, Springer Verlag (1983), 171189.
[7] M. Dugas, P. Hill and K.M. Rangaswamy, Innite rank Butler groups II, Trans. Amer. Math.
Soc., 320, (1990), 643664.
[8] M. Dugas and B. Thome, The functor Bext under the negation of CH, Forum Math. 3 (1991),
2333.
[9] P.C. Eklof and A.H. Mekler, Almost Free Modules - Set-Theoretic Methods, North Holland
Mathematical Library, 46, (1990).
[10] P. Erd os and R. Rado, A partition calculus in set theory, Bull. Amer. Math. Soc., 62, (1956),
427489.
[11] L. Fuchs, Innite Abelian Groups, Vol. I and II, Academic Press (1970 and 1973).
[12] L. Fuchs, A survey on Butler groups of innite rank, Contemp. Math. 171 (1994), 121139).
[13] L. Fuchs, Butler groups of innite rank, J. Pure Appl. Algebra, 98, (1995), 2544.
[14] L. Fuchs and M. Magidor, Butler groups of arbitrary cardinality, Israel J. Math. 84 (1993),
239263.
[15] L. Fuchs and K.M. Rangaswamy, Butler groups that are unions of subgroups with countable
typesets, Arch. Math. 61 (1993), 105110.
[16] R. Hunter, Balanced subgroups of abelian groups, Trans. of the American Math. Soc. 215 (1976),
8198.
[17] T. Jech, Set Theory, Academic Press, New York (1973).
7
5
4


r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
0
-
3
0







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


IT IS CONSISTENT WITH ZFC THAT B1-GROUPS ARE NOT B2 19
[18] K. Kunen, Set Theory - An Introduction to Independent Proofs, Studies in Logic and the Foun-
dations of Mathematics, North Holland, 102 (1980).
[19] M. Magidor and S. Shelah, Bext
2
(G, T) can be non trivial even assuming GCH, Contemp.
Math. 171 (1994), 287294.
[20] K.M. Rangaswamy, A homological characterization of abelian B
2
-groups, Proc. Amer. Math.
Soc. 121 (1994), 409415.
[21] S. Shelah, Proper and Improper Forcing, Perspectives in Mathematical Logic, Springer Verlag
(1998).
Department of Mathematics, Hebrew University, 91904 Jerusalem, Israel, and Rut-
gers University, Newbrunswick, NJ U.S.A., e-Mail: Shelah@math.huji.ac.il,
Department of Mathematics, Hebrew University, 91904 Jerusalem, Israel, e-mail:
lutz@math.huji.ac.il