Professional Documents
Culture Documents
Paul C. Eklof and Saharon Shelah - Absolutely Rigid Systems and Absolutely Indecomposable Groups
Paul C. Eklof and Saharon Shelah - Absolutely Rigid Systems and Absolutely Indecomposable Groups
7
8
r
e
v
i
s
i
o
n
:
2
0
0
0
-
1
0
-
0
6
m
o
d
i
f
i
e
d
:
2
0
0
4
-
0
1
-
1
1
Absolutely Rigid Systems and Absolutely
Indecomposable Groups
Paul C. Eklof
and A
are
not L
and A
do not become
isomorphic in any generic extension of the universe. In that case we say that A
and A
: < 2
of groups of cardinality
, where is less than the rst strongly inaccessible cardinal. But he pointed
out that other constructions, such as Fuchs construction [5] of a rigid system
of groups of cardinality the rst measurable or Shelahs [13] for an arbitrary
cardinal involve non-absolute notions like direct products or stationary sets; so
the rigid systems constructed may not remain rigid when the universe of sets is
expanded. The same comment applies to any construction based on a version
of the Black Box.
Here we show that there do not exist arbitrarily large absolutely rigid sys-
tems. The cardinal () in the following theorem is dened in section 2; it is an
inaccessible cardinal much larger than the rst inaccessible, but small enough
to be consistent with the Axiom of Constructibility.
Theorem 1 If is a cardinal () and A
.
This cardinal () (called the rst beautiful cardinal by the second author
in [14]) is the precise dividing line:
Theorem 2 If is a cardinal < () and is any cardinal (), there is
a family A
, A
) = 0.
Despite the limitation imposed by Theorem 1, the construction used to prove
Theorem 2 yields the existence of a proper class of absolutely dierent torsion-
free groups, in the following strong form. This answers the question of Nadel
in the armative, and also provides a new proof of the existence of arbitrarily
large indecomposables.
Theorem 3 For each uncountable cardinal , there exist 2
torsion-free abso-
lutely indecomposable groups H
i,
: i < 2
) and disjunction (
), and under
existential and universal quantication (x, x). Rather than give formal de-
nitions of other model-theoretic concepts, we will illustrate them with examples.
Thus the formula (y) :
xz(2z = x) (z(3z = y))
is a formula of L
n1
z(p
n
z = y)) (x ,= 0)
is a formula of L
B to
mean that every sentence of L
B
is given by the following ([8], or see [1, pp. 13f]):
Lemma 5 A
P such that p p
and a dom(p
) (resp. b rge(p
)).
It is an easy consequence that if A and B are countable, then A
B if
and only if A
= B. Also, this implies that if it is true in V that A
B,
then A
is , that is, p
1
,=
2
, the groups A
1
and A
2
are not even potentially isomorphic: this is
because for any there is a formula
(x) x ,= 0) is true
in a p-group A if and only if A has length . Indeed, we dene, by induction,
to be
y(py = x
(y))
if = + 1 and if is a limit ordinal dene
to be
<
y(py = x
(y)).
Thus there is a proper class of pairwise absolutely non-isomorphic p-groups.
Although there is not available a standard group-theoretic notion which will
serve the same purpose for torsion-free groups, we will prove in section 5 that
there is a proper class of indecomposable torsion-free abelian groups H
:
a cardinal such that for any ,= , the groups H
and H
are not L
-
equivalent.
2 Quasi-well-orderings and beautiful cardinals
A quasi-order Q is a pair (Q,
Q
) where
Q
is a reexive and transitive binary
relation on Q. There is an extensive theory of well-orderings of quasi-orders
developed by Higman, Kruskal, Nash-Williams and Laver among others (cf.
[12], [9]). A generalization to uncountable cardinals is due to the second author
([14]). The key notion that we need is the following: for an innite cardinal
4
6
7
8
r
e
v
i
s
i
o
n
:
2
0
0
0
-
1
0
-
0
6
m
o
d
i
f
i
e
d
:
2
0
0
4
-
0
1
-
1
1
, Q is called narrow if there is no antichain in Q of size , i.e., for every
f : Q there exist ,= such that f()
Q
f(). (Note that this use of the
terminology antichain in [10, p.32] for example is dierent from its use
in forcing theory.)
A tree is a partially-ordered set (T, ) such that for all t T, pred(t)= s
T : s < t is a well-ordered set; moreover, there is only one element r of T,
called the root of T, such that pred(r) is empty. The order-type of pred(t) is
called the height of t, denoted ht(t); the height of T is supht(t) + 1 : t T.
If Q is a quasi-order, a Q-labeled tree is a pair (T,
T
) consisting of a tree T of
height and a function
T
: T Q. On any set of Q-labeled trees we dene
a quasi-order by: (T
1
,
1
) _ (T
2
,
2
) if and only if there is a function : T
1
T
2
which preserves the tree-order (i.e. t
T1
t
implies (t)
T2
(t
)) as well as
the height of elements and also is such that for all t T
1
,
1
(t)
Q
2
((t)).
One result from [14] that we will use implies that for suciently large cardi-
nals and suciently small Q, any set of Q-labeled trees is -narrow. In order
to state the result precisely we need to dene a certain (relatively small) large
cardinal.
Let () be the rst -Erd os cardinal, i.e., the least cardinal such that
()
<
; in other words, the least cardinal such that for every function F
from the nite subsets of to 2 there is an innite subset X of such that
there is a function c : 2 such for every nite subset Y of X, F(Y ) = c([Y [).
It has been shown that this cardinal is strongly inaccessible (cf. [7, p. 392]).
Thus it cannot be proved in ZFC that () exists (or even that its existence is
consistent). If it exists, there are many weakly compact cardinals below it, and,
on the other hand, it is less than the rst measurable cardinal (if such exists).
Moreover, if it is consistent with ZFC that there is such a cardinal, then it is
consistent with ZFC + V = L that there is such a cardinal ([15]). If () does
not exist, then Theorem 6 is uninteresting. On the other hand, Theorem 7 then
applies to every cardinal , and its consequences, given in section 4, are still of
interest.
The following is a consequence of results proved in [14] (cf. Theorem 5.3, p.
208 and Theorem 2.10, p. 197):
Theorem 6 If Q is a quasi-order of cardinality < (), and o is a set of
Q-labeled trees, then o is ()-narrow.
On the other hand, it follows from results in [14] that for any cardinal smaller
than (), there is an absolute antichain of that size:
Theorem 7 If < (), there is a family T = (T
) : < of -labeled
trees, each of cardinality < (), such that in any generic extension of V , for
all ,= , (T
) (T
).
Some commentary is needed on the absoluteness of the antichain T =
(T
: < is
a family of non-zero abelian groups, where we can assume that = (). For
each < , let T
for some n
s
and s t if and
only if n
s
n
t
and t n
s
= s.
Let Q
ab
be the set of all quantier-free n-types of abelian groups; that is,
Y Q
ab
if and only if for some abelian group G, some n , and some function
s : n G, Y is the set tp
qf
(s/G) of all quantier-free formulas (x
0
, ..., x
n1
)
of L
: T
Q
ab
by letting
(s) = tp
qf
(s/A
) _ (T
), say : T
is
such that s t implies (s) (t) and for all s T
(s)
((s)).
Now move to a generic extension V [G] in which A
is countable. In V [G],
let : A
by
letting f((n)) = ( n + 1)(n) for all n < . To see that f is an embedding,
note that f((n)) = ( k)(n) for all k > n since preserves the tree ordering;
moreover, for any a, b, c A
( k)
(( k)) =
(e.g. a ,= 0, ab = c, ab =
c) is satised by f(a), f(b), f(c) in A
to a
. This proves
Theorem 4.
Ernest Schimmerling has pointed out that there is a soft proof of these
results (not relying on Theorem 6) using a model of set theory (with A
:
< as additional predicate) and a set of indiscernibles given by the dening
property of = ().
4 Existence theorem
In this section we will prove Theorem 2. So let < () and let be a cardinal
(). Let (T
rge(g
n
), and let Y
0
= g
0
(0).
For < , let W
z
: Y
n1
, z Z
,
B
n,
= b
,
: Y
n1
, T
<>
and /
0
= a
0
= B
0,
. We are going to dene A
to be a subgroup of W
.
Since is xed throughout the construction, we will usually omit the subscript
from what follows (until we come to consider Hom(A
, A
)).
For each n > 0, let h
n
be a bijection from /
n
B
n
onto ; let h
0
(a
0
) = 0.
Then for any w /
n
B
n
, and any g
n
(h
n
(w)), we will use a
<>
or b
<>
as
a notation for w. (So a
<>
= b
<>
; moreover, a
<>
= a
<>
if and only if and
belong to the same member of the range of g
n
.) Now we can dene A (= A
)
to be the subgroup of W generated (as abelian group) by the union of
n0
1
p
k
n,m,0
a
z
: m, k , z Z
<>, Y
n
, dom(z) = m (1)
n0
1
p
k
n,m,1
(a
z
+a
zm1
):m, k 0, z Z
, Y
n
, dom(z) = m (2)
7
6
7
8
r
e
v
i
s
i
o
n
:
2
0
0
0
-
1
0
-
0
6
m
o
d
i
f
i
e
d
:
2
0
0
4
-
0
1
-
1
1
and
n0
1
q
k
n,m,
(b
+b
m1
):m, k 0, T
, Y
n
, dom() = m,
() = .
(3)
(where b
1
= 0). We will use the sets (1) and (2) to prove that (I) A
is
absolutely indecomposable and the set (3) to prove that (II) Hom(A
, A
) = 0
for all ,= .
If x A
, we will write p
such that
p
k
v = x. For example, if w /
n
B
n
and g
n
(h
n
(w)) and
(<>) =
0
,
then p
n,0,0
[w and q
n,0,o
[w. Assertions about divisibility in A
n
/
n
B
n
of W
; for example, p
n,m,0
[x if and only if x =
r
i=1
c
i
a
i
zi
for some
1
, ...,
r
in Y
n
, z
i
of length m, and c
i
Q (with denominator a power
of p
n,m,0
).
Proof of (I)
We will show, in fact, that in any generic extension V [G] the only automor-
phisms of A (= A
) are the trivial ones, 1 and 1. This part of the proof does
not use the trees in T ; the absoluteness is a consequence of an argument using
formulas of L
(n , Y
n
) such that
for any u A, A [=
n,
[u] if and only if there are w
1
, ..., w
r
/
n
B
n
, and c
1
, ..., c
r
Z 0, such that u =
r
i=1
c
i
w
i
and
r
i=1
g
n
(h
n
(w
i
)).
Assuming the claim for now, suppose that in V [G] there is an automorphism
F of A. For any n , consider any w /
n
B
n
; since w = a
<>
for
g
n
(h
n
(w)), p
n,0,0
[w; therefore p
n,0,0
[F(w), and hence F(w) =
r
i=1
c
i
w
i
for some distinct w
i
/
n
B
n
. Moreover, by (1A), A [=
n,
[w] if and only
if g
n
(h
n
(w)) if and only if A [=
n,
[F(w)]. Thus, since the elements of
the range of g
n
are disjoint, we must have that r = 1 and w
1
= w, that is,
F(w) = cw for some c = c(w) Q.
If we can show that c(w) = c(a
0
) for all w
n
/
n
B
n
, then F is
multiplication by c(a
0
), and then it is easy to see that c(a
0
) must be 1. It
will be enough to show that if w = a
z
(resp. w = b
) for some Y
n1
, then
c(w) = c(a
<>
), for then c(w) = c(w
) for some w
/
n1
B
n1
(namely, the
unique w
such that g
n1
(h
n1
(w
) = c(a
0
).
So suppose w = a
z
; the proof will be by induction on the length of z that
c(a
z
) = c(a
<>
). Suppose that the length of z = m > 0. Let c = c(a
z
)
8
6
7
8
r
e
v
i
s
i
o
n
:
2
0
0
0
-
1
0
-
0
6
m
o
d
i
f
i
e
d
:
2
0
0
4
-
0
1
-
1
1
and c
= c(a
zm1
). By induction it is enough to prove that c = c
. Since
p
n,m,1
[(a
z
+a
zm1
), it is also the case that p
n,m,1
divides
F(a
z
) +F(a
zm1
) = ca
z
+c
zm1
= c(a
z
+a
zm1
) + (c
c)a
zm1
so p
n,m,1
[(c
c)a
zm1
, which is impossible unless c = c
.
The proof is similar if w = b
. First,
we will use p
[x as an abbreviation for
k
v
k
(p
k
v
k
= x).
Dene
n,m,0
(y) to be p
n,m,0
[y. Then for u A, A [=
n,m,0
[u] if and only
if u is in the subgroup (Z-submodule) generated by
1
p
k
n,m,0
a
z
: Y
n
, k
, z Z
, dom(z) = m. Dene
n,m,
(y) for each m > 0 by recursion on : if
= + 1,
n,m,
(y) is
n,m,
(y) y
(
n,m+1,
(y
) (p
n,m+1,1
[(y +y
)).
If is a limit ordinal, let
n,m,
(y) be
<
n,m,
(y).
Then for u A, A [=
n,m,
[u] if and only if u is in the subgroup generated by
a
z
: Y
n
, z Z
<>
: Y
n
, .
Now dene
n,
(x) to be
n,0,0
(x) y[p
n,1,1
[(x +y)
n,1,
(y)
n,1,+1
(y)].
If u =
r
i=1
c
i
w
i
, for some w
i
/
n
B
n
, then p
n,1,1
[(u+y) i y =
r
i=1
c
i
a
i
<i>
for some
i
g
n
(h
n
(w
i
)); using the conality of members of the range of g
n
, it
follows easily that
n,
(x) has the desired property.
Proof of (II)
Suppose that there is a non-zero homomorphismH : A
. We are going
to use H to dene : T
showing that (T
) _ (T
), contrary to
the choice of the family of labeled trees. Now H(w) ,= 0 for some w /
n
B
n,
9
6
7
8
r
e
v
i
s
i
o
n
:
2
0
0
0
-
1
0
-
0
6
m
o
d
i
f
i
e
d
:
2
0
0
4
-
0
1
-
1
1
for some n . Thus for some Y
n
, H(b
<>,
) ,= 0. Fix such an (which can
in fact be any member of g
n
(h
n
(w))). Let
(<>) =
0
. Then q
n,0,0,0
[b
<>,
so q
n,0,0,0
[H (b
<>,
), and hence H(b
<>,
) must be of the form
r
i=1
c
i
b
i
<>,
where c
i
Q 0, and
(<>) =
0
. So letting (<>) = <> (as it must),
we have conrmed that
(()) =
() for =<>.
Now suppose that for some m 0, () has been dened for all nodes
of T
(()) =
(),
in H(b
,
) is non-zero. Now
consider any node of T
, for =
(),
q
n,m+1,,1
[b
,
+b
,
so q
n,m+1,,1
[H (b
,
)+H (b
,
) in A
(),
in H(b
,
) is non-zero, there must be a node
in T
of
height m + 1 such that
,
in H(b
,
) is
c, and moreover such that
) = . So we can let () =
. This completes
the proof of Theorem 2.
5 Absolutely non-isomorphic indecomposables
In this section we sketch how to modify the construction in the preceding section
in order to prove Theorem 3. (Note that this construction does not require the
trees T
S
i,
: i < 2
be a list of 2
,
dene g
1,i
: T([, + ))) as before but with the additional stipulation
that for all < , g
1,i
() + : S
i,
. (Here again the operation
is ordinal addition.) Let Y
1,i
=
rge(g
1,i
) we will also choose g
1,i
such that
Y
1,i
= + : S
i,
. For convenience, for n ,= 1 we let Y
n,i
denote Y
n
(independent of i).
For each n > 0, let h
n,i
be a bijection from a
z
: Y
n1,i
, z Z
onto
; use these bijections to make identications as in the previous construction.
Then H
i,
is dened to be the subgroup of the Q-vector space with basis
a
0
a
z
: n > 0, Y
n1,i
, z Z
n0
1
p
k
n,m,0
a
z
: m, k , z Z
<>, Y
n,i
, dom(z) = m
and
n0
1
p
k
n,m,1
(a
z
+a
zm1
):m, k 0, z Z
, Y
n,i
, dom(z) = m.
10
6
7
8
r
e
v
i
s
i
o
n
:
2
0
0
0
-
1
0
-
0
6
m
o
d
i
f
i
e
d
:
2
0
0
4
-
0
1
-
1
1
As before, the groups H
i,
are absolutely indecomposable. It remains to
show that for ,= or i ,= j, H
i,
and H
j,
are not L