Professional Documents
Culture Documents
2
0
r
e
v
i
s
i
o
n
:
2
0
0
9
-
0
6
-
0
7
m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
6
-
0
7
Result.Math. 99 (9999), 112
1422-6383/990001-12, DOI 10.1007/s00025-006-xxxx-y
c 2009 Birkh auser Verlag Basel/Switzerland
Results in Mathematics
n
-Free Modules With Trivial Duals
R udiger Gobel and Saharon Shelah
Abstract. In the rst part of this paper we introduce a simplied version of a
new Black Box from Shelah [11] which can be used to construct complicated
n-free abelian groups for any natural number n N. In the second part
we apply this prediction principle to derive for many commutative rings R
the existence of n-free R-modules M with trivial dual M
= 0, where M
=
Hom(M, R). The minimal size of the n-free abelian groups constructed below
is n, and this lower bound is also necessary as can be seen immediately if
we apply GCH.
Mathematics Subject Classication (2000). Primary 13C05, 13C10, 13C13,
20K20, 20K25, 20K30; Secondary 03E05, 03E35.
Keywords. prediction principles, almost free abelian groups, dual groups.
1. Introduction
The existence of almost free R-modules M over countable principal ideal domains
(but not elds) with trivial dual M
= 0 of cardinality [ M [ =
n
for any natural number n;
see Theorem 4.3 and Corollary 4.4. It remains an open question if we can go any
further and pass
or possibly replace M
2
+1
-free implies
2
+2
-free.
We would like to thank Daniel Herden for several very useful suggestions and
some corrections.
2. The Combinatorial Black Box for
The new Black Box depends on a nite sequence of cardinals satisfying some
cardinal conditions. Thus let k
< and =
1
, . . . ,
k
) be a sequence of
cardinals such that for
l
:=
0
l
(l k
l
l+1
=
l+1
(l < k
). (2.1)
We will also say that is a -sequence and note that
l
=
0
l
<
l+1
.
Condition (2.1) is used to enumerate all maps which we want to predict before
constructing the modules. If is any cardinal, then we can dene inductively a
-sequence: Let
1
=
0
and if
l
is dened for l < k
l+1
>
l
with (2.1), e.g. put
l+1
=
l
l
. The sequence
1
, . . . ,
k
) is an example
of such a -sequence.
If is a cardinal, then
will denote all order preserving maps :
(which we also call innite branches) on , while
>
denotes the family of all
order preserving nite branches : n on , where the natural number n,
and (the rst innite ordinal) are considered as sets, e.g. n = 0, . . . , n 1,
thus the nite branch has length n.
9
2
0
r
e
v
i
s
i
o
n
:
2
0
0
9
-
0
6
-
0
7
m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
6
-
0
7
Vol. 99 (9999) n-free modules with trivial duals 3
For the sake of generality we rst consider any sequence =
1
, . . . ,
k
)
of cardinals such that
0
l
=
l
(l k
. Thus let
=
k
.
For the second set we replace the m-th (and only the m-th) coordinate
m
by
the nite branches
>
m
, thus
m
=
1
>
k
for m k
and let
=
_
mk
m
. (2.2)
The elements of ,
l
or
l
>
, respectively. Using these s as support of elements of
the module will make enough room for linear independence which will then give
n
-freeness.
With each member of we can associate a subset of
:
Denition 2.1. If = (
1
, . . . ,
k
) and m k
)
( m, n))
l
=
_
l
if m ,= l k
m
n if l = m.
We associate with its support [] = m, n) [ m k
, n < which is a
countable subset of
. Similarly, for m k
m
for all m k
. We let C =
mk
C
m
and dene a set-trap (for , C) as a
map
: [] C with a label .
The following lemma will be used for the inductive proof of our next theorem.
Lemma 2.3. Let be an innite cardinal, =
0
and P a set of size [ P[ = .
Then there is a sequence
[
) such that
(a)
=
n
[ n < ), with
n
P,
(b) If f = f
[ f
P,
>
, and
>
, then there is
[, we can x an embedding
: P
.
And since [
>
[ = there is also a list
>
=
= is unbounded.
Moreover we dene for each n < a coding map
n
:
n
P
n
2
>
9
2
0
r
e
v
i
s
i
o
n
:
2
0
0
9
-
0
6
-
0
7
m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
6
-
0
7
4 R udiger Gobel and Saharon Shelah Result.Math.
=
0
, . . . ,
n1
)
n
= (
0
n)
. . .
(
n1
n).
Finally let X
be the collection of all order preserving maps : such
that the following holds.
=
i
[ i < )
P and < with ( n)
n
=
n
for all n > k. (2.3)
By denition of
n
it follows that is uniquely determined by (2.3). (Just
note that,
n
determines
m
n for all n > k, m.)
We now prove the two statements of the lemma. For (a) we consider any
. If / X, then we can choose arbitrary members
n
P, and if X,
then choose the uniquely determined sequence from (2.3) and let
n
=
n
, so
= .
For (b) we consider some f = f
[ f
P,
>
and
>
. In
this case we must dene an extension =
n
[ n < )
of . Thus put
0 = ,
n
= n for n < lg(). And if n lg(), then using the above and
that the list of
as
above and C as in Denition 2.2. Then there is a family of set-traps
[ )
satisfying the following
prediction principle: If :
) and
k
, then for some there is a set-trap
with
and 0
k
= .
Proof. The proof of Theorem 2.4 will follow by induction on k
. Temporarily
we will attach parameters k
,
k
, . . . .
The rst step is k
k
and
k
=
>
k
and m, n) =
k
n holds. We
put P = C
k
= C
k
k
and note that [ P[
k
=
0
k
. The trap functions
are
dened by
( m, n))
k
n
and with f
k
n
= f
k
n
= ( m, n))
and the prediction principle in Theorem 2.4 is clear.
The induction step k
= k + 1:
Suppose that Theorem 2.4 is shown for k. We must nd a family of traps
k
[
9
2
0
r
e
v
i
s
i
o
n
:
2
0
0
9
-
0
6
-
0
7
m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
6
-
0
7
Vol. 99 (9999) n-free modules with trivial duals 5
k
for
k
, C
k
and verify the prediction principle in Theorem 2.4. By induction
hypothesis there is such a family
k
[
k
for
k
, C
k
.
Let
k
= ,
k
= and recall that =
0
. Moreover, by assump-
tion
C
k
C
k
k
[
k
[
k
= by condition (2.1) of the -sequence.
If
k
, then let
k
: [] C
k
be the following map in (2.4). Recall that
=
1
, . . . ,
k
) and thus
k
and
k
n
map(
k
, C
k
k
) by Lemma 2.3.
If
=
1
, . . . ,
k
)
k
, then
k
n
is a well-dened element of C
k
k
and
k
is
given by induction hypothesis. We can now dene
( m, n))
k
=
_
k
n
if m = k
m, n))
k
if m < k
.
(2.4)
In order to show the prediction principle we consider an arbitrary map :
C
k
satisfying the trap-condition
m
C
m
for all m k
. We want to
nd
k
and
such that
)
k
satises [
] =
k
and
0
k
= (which is the claim of Theorem 2.4).
First we search for and dene for each
>
a map f
:
k
C
k
k
from P depending on . If
k
, then
)
k
, thus
f
:= (
)) (2.5)
is well-dened. By the Lemma 2.3 we nd
such that
0 = and f
n
=
n
:
k
C
k
k
for all n .
By Lemma 2.3(b), (2.4) and (2.5) we have for any and
) that
(
, n))
k
=
n
= f
n
= (
n)) = (
, n))
and 0 = which is the prediction as required for m = k
.
Now we consider the case when m < k
:
k
C
k
depending on and . If
, then
)
k
(because
), thus
:= (
))
is well-dened and by induction hypothesis on the traps
k
there is some
k
such that
( m, n))
k
= ( m, n))
)
k
. By the last displayed equation and (2.4) we have
for m < k
that
(
m, n))
k
= ( m, n))
k
= ( m, n))
= (
m, n)).
Thus
k
predicts with 0
k
= 0 = as suggested above.
9
2
0
r
e
v
i
s
i
o
n
:
2
0
0
9
-
0
6
-
0
7
m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
6
-
0
7
6 R udiger Gobel and Saharon Shelah Result.Math.
Denition 2.5. Let F :
[ <
) of (we write
[ < )
and there are
, n
< ( <
and n
, n) /
, n) [ <
F.
Moreover, is -free (with respect to F) for some cardinal if the above holds for
all subsets of of cardinality < .
This is to say, that every newly chosen element
,
of cardinality [ [
k
and u
1, . . . , k
[ [ u
[ > k, ) we can
nd an enumeration
[ <
k
) of ,
and n
< ( <
k
) such that
, n) /
, n) [ <
F for all n n
.
Proof. The proof follows by induction on k. We begin with k = 0, hence we
may assume that [ [ =
0
. Let =
[ follows u
for
all < . To be denite we may choose
= min u
. If ,= < , then
,=
and there is n
,
such that
, n) , =
, n) for all n n
. Since
, n) /
F for
all n n
,
. If n
= max
<
n
,
, then
, n) /
, n) [ <
F
for all n n
<
k
with
0
= and
[
1
[ =
k
. The crucial idea comes from [11]: We can also assume that this chain
is closed, meaning that for any <
k
, ,
and with
m
[ m k
m
,
m
, (F)
m
, (
F)
m
[ m k
follows
. Thus, if
+1
= k
[ n < ,
such that , n) = , n) or , n) = F
is empty or a singleton. Otherwise there are n, n
with
, n) , n), F and
, n
, n
),
F for certain ,
.
Hence
m
[ m k
m
,
m
, (F)
m
, (
F)
m
[ m k
.
If <
k
, then let D
=
+1
and u
:= u
( <
k
) of D
[ <
k
)
of satisfying the proposition.
9
2
0
r
e
v
i
s
i
o
n
:
2
0
0
9
-
0
6
-
0
7
m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
6
-
0
7
Vol. 99 (9999) n-free modules with trivial duals 7
3. The Black Box for
n
-free modules
Let R be a commutative ring with S a countable multiplicatively closed subset
such that the following holds.
(i) The elements of S are not zero-divisors, i.e. if s S, r R and sr = 0, then
r = 0.
(ii)
sS
sR = 0.
We also say that R is an S-ring. If (i) holds, then R is S-torsion-free and if (ii) holds,
then R is S-reduced, see [8]. To ease notations we use the letter S only if we want
to emphasize that the argument depends on it. If M is an R-module, then these
denitions naturally carry over to M. Finally we enumerate S = s
n
[ n < , let
s
0
= 1 and put q
n
=
in
s
i
, thus q
n+1
= q
n
s
n+1
.
Similar to the Black Box in [1], we rst dene the basic R-module B, which
is
B =
Re
.
Denition 3.1. If U
U
Re
of
B, and in particular, let B
= B
[]
and B
m
= B
[ m]
be the canonical summand
of and m ( ), respectively.
We have several Rfree summands
B
m
=
n<
Re
m,n
and B
mk
B
m
.
The S-topology (generated by the basis sB (s S)) of neighbourhoods of 0 is
Hausdor on B and (as usual) we can consider the S-completion
B of B; see [8] for
elementary facts on the elements of
B. Let
B =
Re
. Every element b
B
has a natural
-support [b]
which contribute to
the sum-representation b =
with coecients 0 ,= b
R. Thus let
[b]
[ b
is nite. As in the earlier Black Boxes (see [8]) we use conditions on the support
(given by the prediction) to select (carefully) elements from
B added to B to get
the nal structure M, such that
B M
B
which is an S-pure submodule of
B, thus satisfying
M s
B sM
for all s S. Thus S-topological arguments can be used carelessly switching be-
tween these three modules.
We will now use B,
: Dom(
) G, such that B
Dom(
) B.
The -Black Box 3.3. Given a -sequence =
1
, . . . ,
k
) with (2.1) and an
R-module G of size [ G[
1
, let ,
.
Proof. The theorem is an immediate consequence of Theorem 2.4. We view
the set maps in Theorem 2.4 as the restrictions of the R-homomorphisms in The-
orem 3.3 to the canonical R-basis e
of B. There is a one-to-one
correspondence between these maps and thus Theorem 3.3 follows.
4. The R-modules
Let R be an S-torsion-free and S-reduced commutative ring of size [ R[ < 2
0
,
k
=
0
k
=
k
be as before, B =
Re
and
=
_
[] with [] = m, n) [ m k
, n < .
We also choose any bijection
:
k
.
Thus we can write the basis elements of B in the form e
()
for any
k
.
From [5] follows that the S-adic completion
R of R has 2
0
algebraically
independent elements over R, and in particular
= 2
0
.
Next we dene particular elements in
B. If , then let
y
k
=
nk
q
n
q
k
(
k
m=1
e
m,n
+b
n
e
(0
k
)
)
where b
n
R. Moreover let y
= y
0
. We will choose
R and write
n<
q
n
b
n
and let
k
=
nk
qn
q
k
b
n
. Thus
y
k
=
nk
q
n
q
k
(
k
m=1
e
m,n
) +
k
e
(0
k
)
and from
s
k+1
y
k+1
=
nk+1
q
n
q
k
(
k
m=1
e
m,n
+b
n
e
(0
k
)
)
9
2
0
r
e
v
i
s
i
o
n
:
2
0
0
9
-
0
6
-
0
7
m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
6
-
0
7
Vol. 99 (9999) n-free modules with trivial duals 9
and y
k
s
k+1
y
k+1
=
k
m=1
e
m,k
+b
k
e
(0
k
)
, follows
s
k+1
y
k+1
= y
k
k
m=1
e
m,k
b
k
e
(0
k
)
. (4.1)
We want to dene an R-module M with B M
B which is S-pure in
B. Thus M/B is S-torsion-free and S-divisible. It follows that for any non-trivial
homomorphism : M R there is
with e
,= 0. If , then we will
adjoin to B for some suitable
R the element y
n<
q
n
(
k
m=1
e
m,n
)+
e
(0
k
)
. This will follow with the help of the next
Proposition 4.1. Let R be an S-torsion-free and S-reduced commutative ring of size
< 2
0
. Then for any there are
R and
y
n<
q
n
(
k
m=1
e
m,n
) +
e
(0
k
)
(4.2)
with no homomorphism : B, y
R such that B
[]
=
and e
(0
k
)
,=
0.
Proof. Let e = e
(0
k
)
and choose pairwise distinct elements
R ( <
2
0
). Moreover let y =
n<
q
n
(
k
m=1
e
m,n
) and put y
= y +
e. Sup-
pose that for each < 2
0
there is a homomorphism
: B, y
R
with
B
[]
=
and e
= y
and also e
= e
=: c ,= 0. But
this implies
0 = y
= (y +
e)
(y +
e)
= y
= (
)c.
And from
,= 0 follows c = 0, a contradiction.
Finally we dene the R-module
M = B, y
[ )
B. (4.3)
Here we let y
-support [g]
[ )
and e
m,n
and simply collect the s and m, n) needed. Clearly [g] is a nite
subset of
= e
m,n
, e
(0
k
)
, y
[ , m k
, n < )
B,
which also has size < . Thus, in order to show
k
-freeness of M, we only must
consider any of size [ [ <
k
and show the freeness of the module M
.
We may assume that [ [ =
k1
. Let F :
.
By Proposition 2.6 we can express the generators of M
of the form
M
= e
m,n
, e
F
, y
n
[ <
k1
, m k
, n < )
and nd a sequence of pairs (
, n
) (k
, n) /
, n) [ <
F [ < . (4.4)
Let M
= e
m,n
, e
F
, y
n
[ < , m k
+e
m,n
, e
F
, y
n
[ m k
, n < )
= M
+e
,n
[ n < n
) +y
n
[ n n
)
+e
F
, e
m,n
[
,= m k
, n < ).
Hence any element in M
+1
/M
can be represented in M
+1
modulo M
of
the form
nn
r
n
y
n
+
n<n
r
n
e
,n
+re
F
+
n<
=mk
r
mn
e
m,n
.
Moreover, the summands involving the e
m,n
s have disjoint supports.
Now condition (4.4) applies recursively. And by the disjointness (identifying e
F
with one of the e
m,n
s if possible) it also follows that all coecients r, r
n
, r
mn
must be zero, showing that the set
e
,k
, e
F
, e
m,n
[ k < n
,= m k
, n < M
freely generates M
+1
/M
. Thus M
is free.
We can nally show the
9
2
0
r
e
v
i
s
i
o
n
:
2
0
0
9
-
0
6
-
0
7
m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
6
-
0
7
Vol. 99 (9999) 11
Theorem 4.3. Let R be an S-torsion-free and S-reduced commutative ring of size
< 2
0
. Then for any -sequence =
1
, . . . ,
k
) with (2.1) there exists an
k
-
free R-module M of size
k
with trivial dual Hom(M, R) = 0. In particular, if R
is a principal ideal domain but not a eld of size < 2
0
, then there is an
k
-free
R-module M of size
k
with trivial dual.
Proof. If M is the R-module above, then M is
k
-free by Proposition 4.2.
Obviously M has size
k
. If : M R is a non-trivial homomorphism, then
there is
n
with trivial dual.
References
[1] A. L. S. Corner and R. Gobel, Prescribing endomorphism algebras A unied treat-
ment, Proc. London Math. Soc. (3) 50 (1985), 447 479.
[2] A. L. S. Corner and R. Gobel, Small almost free modules with prescribed topological
endomorphism rings, Rendiconti Sem. Mat. Univ. Padova 109 (2003) 217 234.
[3] K. Eda, Cardinality restrictions on preradicals. pp. 277283 in Abelian group theory
(Perth, 1987), Contemp. Math. 87, Amer. Math. Soc., Providence, RI, 1989.
[4] P. Eklof and A. Mekler, Almost Free Modules, Set-theoretic Methods, North-Holland,
2002.
[5] R. Gobel and W. May, Independence in completions and endomorphism algebras,
Forum mathematicum 1 (1989), 215 226.
[6] R. Gobel and S. Shelah, G.C.H. implies the existence of many rigid almost free
abelian groups, pp. 253 271 in Abelian Groups and Modules, Marcel Dekker New
York 1996.
[7] R. Gobel and S. Shelah, Indecomposable almost free modules the local case, Cana-
dian J. Math. 50 (1998), 719 738.
[8] R. Gobel and J. Trlifaj, Endomorphism Algebras and Approximations of Modules,
Expositions in Mathematics 41, Walter de Gruyter Verlag, Berlin (2006).
[9] T. Jech, Set Theory, Monographs in Mathematics, Springer, Berlin Heidelberg
(2002).
[10] M. Magidor and S. Shelah, When does almost free imply free? (For groups, transver-
sals, etc.), J. Amer. Math. Soc. 7 (1994), 769830.
[11] S. Shelah, n-free abelian groups with no non-zero homomorphisms to Z, Cubo - A
Mathematical Journal 9 (2007), 59 79.
9
2
0
r
e
v
i
s
i
o
n
:
2
0
0
9
-
0
6
-
0
7
m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
6
-
0
7
12 R udiger Gobel and Saharon Shelah Result.Math.
R udiger Gobel
Fachbereich Mathematik,
Universitat Duisburg Essen
Campus Essen, D 45117 Essen, Germany
e-mail: ruediger.goebel@uni-due.de
Saharon Shelah
Institute of Mathematics,
Hebrew University, Einstein Institute of Mathematics,
Givat Ram, Jerusalem 91904, Israel and
Rutgers University,
Department of Mathematics, Hill Center, Busch Campus
Piscataway, NJ 08854 8019, U.S.A
e-mail: shelah@math.huji.ac.il
Received: September 12th, 2007