POSSIBLE CARDINALITIES OF MAXIMAL ABELIAN

SUBGROUPS OF QUOTIENTS OF PERMUTATION GROUPS OF

THE INTEGERS
SAHARON SHELAH AND JURIS STEPR

ANS
Abstract. If G is a group then the abelian subgroup spectrum of G is dened to
be the set of all such that there is a maximal abelian subgroup of G of size . The
cardinal invariant A(G) is dened to be the least uncountable cardinal in the abelian
subgroup spectrum of G. The value of A(G) is examined for various groups G which
are quotients of certain permutation groups on the integers. An important special
case, to which much of the paper is devoted, is the quotient of the full symmetric
group by the normal subgroup of permutations with nite support. It is shown that,
using G to denote this group, A(G) a. Moreover, it is consistent that A(G) ,= a.
Related results are obtained for other quotients using Borel ideals.
1. Introduction and Definitions
The maximality of abelian subgroups plays a role in various parts of group theory.
For example, Mycielski [9, 8] has extended a classical result of Lie groups and shown
that a maximal abelian subgroup of a compact connected group is connected. For
nite symmetric groups the question of the size of maximal abelian subgroups has
been examined by Burns and Goldsmith in [4] and Winkler in [16]. It will be shown
in Corollary 3.1 that there is not much interest in generalizing this study to innite
symmetric groups; the cardinality of any maximal abelian subgroup of the symmetric
group of the integers is 2

0
. The purpose of this paper is to examine the size of maximal
abelian subgroups for a class of groups closely related to the the symmetric group of
the integers; these arise by taking an ideal on the integers, considering the subgroup of
all permutations which respect the ideal and then taking the quotient by the normal
subgroup of permutations which x all integers except for a set in the ideal. It will be
shown that the size of maximal abelian subgroups in such groups is sensitive to the
nature of the ideal as well as to various set theoretic hypotheses.
The reader familiar with applications of the Axiom of Choice may not be surprised
by the assertion just made since one can imagine constructing ideals on the integers
by transnite induction such that the quotient group just described exhibits various
1991 Mathematics Subject Classication. 03E17, 03E35, 03E40, 03E50, 20B07, 20B30, 20B35.
The research of the rst author was supported by The Israel Science Foundation founded by the
Israel Academy of Sciences and Humanities, and by NSF grant No. NSF-DMS97-04477. Research of
the second author for this paper was supported by the Natural Sciences and Engineering Research
Council of Canada. This is number 786 on the rst authors personal list of publications.
1
2 S. SHELAH AND J. STEPR

ANS
desired properties. Consequently, it is of interest to restrict attention to only those
ideals which do not require the Axiom of Choice for their denition. All of the ideals
considered will have simple denitions indeed, they will all be Borel subsets of
T() with the usual topology and, in fact, the rst three sections will focus on
the ideal of nite sets. It should be mentioned that there is a large body of work
examining the analogous quotients of the Boolean algebra T() modulo an analytic
ideal the monograph [6] by Farah is a good reference for this subject. However, the
analogy is far from perfect since, for example, whereas the Boolean algebra T()/[]
<
0
can consistently have 2
2

0
automorphisms [10] it is shown in [1] that the quotient of
the full symmetric group of the integers modulo the subgroup of nite permutations
has only countably many outer automorphisms. Nevertheless, it may be possible
to employ methods similar to those of [6] in order to distinguish between dierent
quotient algebras up to isomorphism. This has been done for elementary equivalence
in [12, 14] for quotients of the full symmetric group on by the normal subgroups
xing all but elements. However, since the full symmetric group of the integers
has only two proper normal subgroups [11], quotients of certain, naturally arising
subgroups will be considered instead. One of the goals of this study is to use the
cardinal invariants associated with maximal abelian subgroups as a tool to distinguish
between isomorphism types of such groups.
In order to state the main results precisely some notation is needed.
Denition 1.1. If G is a group then dene the abelian subgroup spectrum of G to be
the set of all such that there is a maximal abelian subgroup of G of size . Dene
A(G) to be least uncountable cardinal in the abelian subgroup spectrum of G.
The requirement that A(G) be uncountable rather than just innite is important
because many groups have maximal abelian subgroups isomorphic to Z. In particular,
quotients of the symmetric group on N often have a xed point free permutation
of N consisting of a single cycle which generates a maximal abelian subgroup. The
question of whether maximal abelian subgroups which are not nitely generated are
uncountable will not be considered here.
Notation 1.1. Throughout this paper the symbol S will be used to denote the sym-
metric group on N. For S let supp() denote the support of which is dened to
be x domain() : (x) ,= x. If 1 is an ideal
1
on N then S(1) S will denote the
subgroup of all permutations preserving 1; in other words, a permutation belongs
to S(1) provided that (A) 1 if and only if A 1. On the other hand, F(1) will be
used to denote the normal subgroup of S(1) consisting of all permutations S(1)
such that supp() 1. The abbreviation F = F([N]
<
0
) will also be used.
1
An ideal is a collection of subsets of the integers closed under nite unions and subsets.
MAXIMAL ABELIAN SUBGROUPS 3
The focus of this paper will be on computing A(S(1)/F(1)) for various simply
dened ideals. This cardinal will be denoted by A(1).
Notation 1.2. Given a pair of permutations ,
t
[S]
2
dene
NC(,
t
) = n N : (
t
(n)) ,=
t
((n)).
A pair of permutations ,
t
[S]
2
will be said to almost commute modulo an ideal
1 if NC(,
t
) 1 and they will be said to almost commute if NC(,
t
) is nite.
Notation 1.3. Given a permutation and X N dene the orbit of X under by
orb

(X) =
i
(x)
iZ,xX
. The abbreviation orb

(n) = orb

(n) will be used when

no confusion is possible. If o is a set of permutations then dene
orb
S
(X) =
_
n

i=1

j
i
(x)
_
n,xX,
i
S,j1,1
.
Notation 1.4. Given a set of permutations o S dene
S
to be the set of all
non-empty, minimal sets closed under the group generated by the permutations in o.
Dene
S
(n) to be the unique element of
S
containing n. If A and B are in
S
then dene A to be o-isomorphic to B if there is a bijection : A B such that
((a)) = ((a)) for each o and a A.
Notation 1.5. Given two nite subsets A and B of N of the same cardinality, dene

A,B
: A B to be the unique order preserving mapping between them and let

A
=
A,[A[
. Let
A,B
=
A,B

B,A
.
The set theoretic notation used throughout will adhere to the contemporary stan-
dard. In particular, [X]
k
will denote the family of subsets of X of cardinality k and
[X]
<k
will denote the family of subsets of X of cardinality less than k. Occasionally

will be used to denote equivalence modulo a nite set. Since cardinal invariants of
the continuum are closely linked to the investigation of A(1) it is worthwhile recalling
the denitions of some well known invariants.
Denition 1.2. Given an ideal 1 T() let T()/1 be the quotient Boolean algebra
and denote the least cardinal of a maximal, uncountable, pairwise disjoint family
2
in
T()/1 by a(1). In the special case 1 = [N]
<
0
the invariant a(1) is denoted by a
and it should be noted that a is also the least cardinal of an innite, maximal almost
disjoint family; namely a family / T() such that any two of its elements have
nite intersection and / is maximal with respect to this property. The least cardinal
of an ideal B T()/[]
<
0
such that there is no C T()/[]
<
0
disjoint from all
members of B (other than the equivalence class of the nite sets) is denoted by p.
2
See [15] for a more detailed discussion of this invariant.
4 S. SHELAH AND J. STEPR

ANS
In Section 2 it is shown that a is an upper bound for A([N]
<
0
) while in Section 3
it is shown that p serves as a lower bound for A([N]
<
0
). Sections 4 and 5 deal with
consistency results. In Section 4 it is shown that a is not the best possible upper
bound for A([N]
<
0
) since in the iterated Laver model A([N]
<
0
) is strictly less than
a. Sections 5 and 6 deal with quotients using ideals other than the ideal of nite sets.
It is shown in Section 5 that adding
1
Cohen reals to a model where 2

0
>
1
yields
a model where A(1
1/x
) =
1
< 2

0
and 1
1/x
is the ideal of sets whose reciprocals form
a series with nite sum. Section 6 deals with ideals similar to the density ideal. It is
shown that A(1) = 2

0
for many of these ideals 1. No extra set theoretic axioms are
used here. The nal section contains some open questions.
2. An upper bound
Proposition 2.1. A([N]
<
0
) a.
Proof. Let / be a maximal almost disjoint family of subsets of N of size a and let
F(/) be the free abelian group generated by /; in other words, F(/) consists of
all f : / Z such that f has nite support. For a / dene
a
: a a by

a
(i) = min(j a : j > i) and, for j Z, let
j
a
denote the j-fold composition of
a
,
noting that both the domain and range of
j
a
are co-nite subsets of a. If f F(/)
then let (f) be the set of all permutations such that there is a nite set F N
such that if a and a
t
are distinct elements of supp(f) then a a
t
F and such that
(j) =
_

f(a)
a
(j) if j a F and f(a) ,= 0
j if j / F and (a /)j / a or f(a) = 0
leaving (f) undened if there are no such permutations. Observe that (0) = F
where 0 denotes the constant function with value 0. Also note that if (f) and
(g) then (f +g). Since it is easy to see that if F and (f) then
both and are in (f), it follows that (f) is a coset of F if it is dened.
Since is easily seen to be one-to-one, it is an isomorphism between a subgroup of
F(/) and the subgroup (F(/)) of S/F.
In fact, (f) is dened precisely when

a,
f(a) = 0. To see this, let f F(/) and
suppose that the support of f is B and

bB
f(b) = 0. Let F N be a nite set such
that bb
t
F for any two distinct b and b
t
in B and such that Fb is an initial segment
of b for each b B. Let B
+
= b B : f(b) > 0 and B

= b B : f(b) < 0.
If b B
+
let b

be the rst f(b) elements of b F and if b B

let b

be the rst
f(b) elements of b F. Let :

bB

bB
+
b

be any bijection and dene as

MAXIMAL ABELIAN SUBGROUPS 5
follows:
(j) =
_

_
j if j /

bB
b F

f(b)
b
if j b B
+

f(b)
b
if j b b

and b B

(j) if j

bB

and observe that is a bijection. Moreover, F

bB

witnesses that (f).

It is an easy exercise to use the maximality of / to show that if

f(a) ,= 0 then
(f) = .
To see that (F(/)) is maximal abelian let /F S/F(F(/)). Before continuing,
some notation will be introduced. Given two distinct elements a and a
t
of / let
f
a,a
F(/) be such that supp(f
a,a
) = a, a
t
and f
a,a
(a) = 1 = f
a,a
(a
t
). Choose

a,a
(f
a,a
).
Claim 1. If a / is such that supp() a is innite then supp() a is a co-nite
subset of a.
Proof. Let a
t
/a. If the claim fails then there are innitely many n a such that
n / supp() but
a,a
(n) supp(). For any such n it follows that
a,a
(n) ,=
a,a
(n)
while
a,a
(n) =
a,a
(n). Hence /F and
a,a
/F do not commute.
Claim 2. If a / is such that supp() a is innite then (a)

a.
Proof. If not, there are innitely many n a such that (n) / a. Let X be the set of all
such n and choose a
t
/a such that (X) a
t
is innite. Then
a,a
(n) = (n)
and
a,a
(n) ,= n for all but nitely many n
1
((X) a
t
). From the last inequality
it follows that (
a,a
(n)) ,= (n) and hence
a,a
/F does not commute with /F.
Claim 3. If a / is such that supp() a is innite then there is some i Z such
that a

i
a
a.
Proof. From Claim 1 and Claim 2 it follows that for almost all n a there is some
k(n) such that (n) =
k(n)
a
(n). Let a
t
/ a. If the claim is false then there
are innitely many n a such that k(n) ,= k(
a
(n)) = k(
a,a
(n)). For any such n it
follows that

a,a
(n) =
k(n)+1
a,a

(n) ,=
k(
a,a
(n))+1
a,a

(n) =
k(
a,a
(n))
a,a

(
a,a
(n)) =
a,a
(n)
and hence
a,a
and
a,a
disagree on innitely many integers.
There are now two cases to consider.
Case One. There is a nite subset a
1
, a
2
, . . . a
n
/ such that supp()

n
i=1
a
i
.
In this case, use Claim 3 to choose integers k
i
Z such that
a
i

k
i
a
i
6 S. SHELAH AND J. STEPR

ANS
for each i n. This contradicts that /F / (F(/)).
Case Two. There is no nite subset a
1
, a
2
, . . . a
n
/ such that supp()

n
i=1
a
i
.
In this case there are uncountably many a / such that supp() a is innite.
Use Claim 3 to conclude that there is some i Z such that, without loss of generality,
a

i
a
for uncountably many a /. Hence there is some k N such that
n a : n k =
i
a
n a : n k for uncountably many a /. Therefore,
there are distinct a and b in / such that n a : n kn b : n k is not empty.
If j is the maximal element of this intersection then (j) =
i
a
(j) ,=
i
b
(j) = (j).
3. A lower bound
The next series or preliminary lemmas will be used in the proof of Theorem 3.2 which
establishes a lower bound for A([N]
<
0
). Corollary 3.1 has as a trivial consequence the
fact that any maximal abelian subgroup of the full symmetric group of the integers has
cardinality 2

0
; however, this can also be shown by using the topology of point-wise
convergence on this group and noting that any maximal abelian subgroup must be
closed and uncountable, and hence have cardinality 2

0
.
Lemma 3.1. Let o be a nite subset of S whose elements almost commute with each
other.
(1) If all the orbits of each o are nite then each element of
S
is nite.
(2) If, in addition, for each o all the orbits of have size less than or equal to
m() then the cardinality of all but nitely many elements of
S
is no greater
than

S
m().
Proof. Proceed by induction on n = [o[, the case n = 1 being trivial. If the lemma is
true for n let o =
i

n+1
i=1
and let o
t
=
i

n
i=1
. Dene
B =
n
_
i=1
NC(
i
,
n+1
)
and, if the orbits of each o are bounded by m() then let B
t
be the union of those
nitely many A
S
whose cardinality is not bounded by

n
i=1
m(
i
). Dene
B

=
_
morb

n+1
(BB

S
(m)
2 Observe that B

is nite by the induction hypothesis and the fact the orbits of

n+1
are nite. Hence, it suces to show that if C
S
and C B

= then
C
t
= orb

n+1
(C) belongs to
S
. The fact that it is nite is immediate from the
hypothesis that all orbits of
n+1
are nite; similarly, if [C[

n
i=1
m(
i
) then it
follows that [C
t
[

n+1
i=1
m(
i
).
MAXIMAL ABELIAN SUBGROUPS 7
To see that C
t

S
it suces to show that if i n and c C
t
then orb

i
(c) C
t
.
There is some d C such that c orb

n+1
(d). Since orb

i
(d) C C
t
it follows
that if orb

i
(c) , C
t
then there must be some e orb

n+1
(d) such that
i
(e) C
t
and

i
(
n+1
(e)) / C
t
. But
n+1
(
i
(e)) C
t
by denition. Hence
n+1

i
(e) ,=
i

n+1
(e)
contradicting that e / B.
Lemma 3.2. Let o S be nite and suppose that S and S almost commute
with each member of o. Then there is a nite set Y such that for any set X, if
XY = XY then orb
S
(X) = orb
S
(X). Moreover, if and actually
commute with each member of o then Y can be taken to be the empty set.
Proof. Let Y
t
=

S
NC(, ) NC(, ) and let Y =

S
(Y
t
)
1
(Y
t
). Note
that orb
S
(X Y ) =

i=0
X
(i)
where X
(0)
= X Y and X
(n+1)
=

S
orb

(X
(n)
)
and, hence, it suces to show by induction that X
(n)
= X
(n)
for each n
assuming that X
(0)
= X
(0)
. To this end, suppose that X
(n)
= X
(n)
and
x X
(n+1)
X
(n)
. Then there is some x X
(n)
and o such that x orb

( x).
But ( x) = ( x) and hence
k
(( x)) =
k
(( x)) for any k. If n > 1 then x / Y and
it follows that (
k
( x)) = (
k
( x)) for all k. Since x =
k
( x) for some k the result
follows.
If n = 1 it will be shown by induction on [k[ that if x X
(1)
and x X
(0)
and
o are such that x =
k
( x) then (x) = (x). If [k[ = 0 this is immediate. First
assume that k > 0 and (
k1
( x)) = (
k1
( x)). If
k1
( x) / Y
t
then
k1
( x) /
NC(, ) NC(, ) and so
(
k
( x)) = ((
k1
( x))) = ((
k1
( x))) = (
k
( x))
as required. On the other hand, if
k1
( x) Y
t
then (
k1
( x)) Y and so (
k
( x)) =
(
k
( x)) in this case also. The case that k < 0 is handled similarly.
Denition 3.1. If H S is a subgroup then dene H to be strongly almost abelian if
and only if for each h H there is a nite set F(h) N such that if h
1
and h
2
belong
to H then NC(h
1
, h
2
) F(h
1
) F(h
2
).
Lemma 3.3. If H S is an uncountable subgroup and F : H [N]
<
0
attests to the
fact that H is strongly almost abelian then there is a perfect set P S and a nite
W N such that:
There is some g

H such that for all n N W and P either (n) = n

or (n) = g

(n).
NC(, h) W F(h) h
1
(W) for P and h H.
Moreover, if H is actually abelian and not just strongly almost abelian then W can
be assumed to be empty and it can be concluded that each P commutes with each
h H.
8 S. SHELAH AND J. STEPR

ANS
Proof. Given X N and a nite W N dene cl
0
W
(X) = X,
cl
1
W
(X) = z N W : (h H)(x X F(h)) z = h(x) and z / F(h
1
) X
and let cl
n+1
W
(X) = cl
1
W
(cl
n
W
(X)) and then, let cl
W
(X) =

i=1
cl
i
W
(X). Observe rst
that it follows from an argument similar to that in Lemma 3.2 that, if F(g
1
) W
and F(g
2
) W and g
1
X = g
2
X then g
1
cl
1
W
(X) = g
2
cl
1
W
(X) and hence,
g
1
cl
W
(X) = g
2
cl
W
(X). If, for every W [N]
<
0
there is some A
W
[N]
<
0
such
that A
W
cl
W
(A
W
) = N then it follows that each g H is determined by its values
on F(g) A
F(g)
. This contradicts that H is uncountable.
Therefore it must be the case that there is some W [N]
<
0
such that cl
W
(A) ,= N
for every A [N]
<
0
. Furthermore, it is possible to choose some nite W
t
W such
that the set of all g H such that F(g) W
t
is uncountable. Observe that cl
W
(A)
cl
W
(A) for any A, so it is possible to select a
i

i=1
N such that cl
W
(a
i
)

i=1
is
an innite family covering N W
t
. Observe that if g H is such that F(g) W
t
and
g cl
W
(a
i
) is the identity for all but nitely many i then g is determined by its
values on W a
i
: (n cl
W
(a
i
)) g(n) ,= n. Hence, there must be some g H
such that F( g) W
t
and g cl
W
(a
i
) is not the identity for innitely many i. Let
Z =
_
i N : (n cl
W
(a
i
)) g(n) ,= n and g
1
(W
t
) cl
W
(a
i
) =
_
.
First notice that it follows from the denition of cl
W
and the inclusion F( g) W
t
that
g(cl
W
(a
i
)) cl
W
(a
i
) W
t
. Hence g cl
W
(a
i
) is a permutation of cl
W
(a
i
)
for each i Z. Therefore, if for each t : Z 2 the function g
t
is dened
g
t
(n) =
_
g(n) if n cl
W
(a
i
) and t(i) = 0
n otherwise.
then g
t
is a permutation of N. It is routine to check that each g
t
(h(n)) = h(g
t
(n))
provided that n / W
t
F(h) h
1
(W
t
).
Corollary 3.1. If H S is an uncountable, maximal strongly almost abelian subgroup
then [H[ = 2

0
.
Proof. The maximality of H implies that it must contain the perfect set of the con-
clusion of Lemma 3.3.
Lemma 3.4. If H is a maximal abelian subgroup of S/F and there are

1
/F,
2
/F, . . .
n
/F H
such that, letting o =
1
,
2
, . . .
n
, the set [a[
a
S
is innite, then [H[ = 2

0
.
Proof. Let A
j
= (
S
[N]
j
) and note that / = A
j

j=2
is an innite family of
pairwise disjoint sets. For each non-empty A
j
/ choose some j

n such that
MAXIMAL ABELIAN SUBGROUPS 9

j
A
j
is dierent from the identity. For each F : / 2 dene

F
(k) =
_

j
(k) if k A
j
and F(A
j
) = 1
k otherwise
and observe that
F
is a bijection and that if F and G dier on an innite set then
so do
F
and
G
. It suces to show that NC(
F
, ) is nite for each H and
F : / 2.
To see that this is so, let H and let j be so large that
_
n
_
i=1
NC(
i
, )
_

_
i=1
A
i
_

j
_
i=1
A
i
.
Hence, if k j then A
k
commutes with
i
A
k
for i n. It suces to show that
A
k
is a permutation of A
k
for k j because, if this were the case, then it would
follow that for m A
k

F
(m) =
k
(m) =
k
(m) =
F
(m)
if F(A
k
) = 1 and that

F
(m) = (m) =
F
(m)
if F(A
k
) = 0. To see that A
k
is a permutation of A
k
let a A
k
. Then is
an o-isomorphism from
S
(a) onto
S
((a)). If (a) A

for some ,= k then

[
S
(a)[ = k ,= [
S
((a))[ and this contradicts that is a bijection.
Denition 3.2. If g S then dene I(g) =

a
g
: [a[ =
0
. For a nite set
o S dene
I

(o) =
_
S
_
mI()

S
(m).
Lemma 3.5. If H S is an uncountable, maximal, almost commuting subgroup of
size less than 2

0
then [N]
<
0
I

(o)
S[H]
<
0
generates a proper ideal.
Proof. If not, let B H and C N be nite sets such that I

(B) C = N. Without
loss of generality it may be assumed that NC(b, b
t
) C for each b and b
t
in B. Let
S = A
B
: AC = . Observe that each set in S is innite. In order to see this,
let A S and note that
A I

(B) =
_
bB
_
mI(b)

B
(m)
since AC = . Hence there is some b and some m I(b) such that A
B
(m) ,= .
But since A
B
it must also be the case that m A. Since b B it follows that
orb
b
(m) A and so A is innite by the denition of I(b).
Moreover, S itself is an innite set since Lemma 3.2 would imply that H is countable
otherwise. To see this let Y be a nite set given by Lemma 3.2 such that if and
10 S. SHELAH AND J. STEPR

ANS
almost commute with each b B and (X Y ) = (X Y ) then orb
B
(X) =
orb
B
(X). Assuming S is nite, choose a nite set X such that X A ,= for all
A S. Then

xX
orb
B
(x) S and so any h H is determined by its values on
X Y C.
With these observations in hand, let S
S
be the symmetric group on S and dene
: H S
S
by (h)(s) = t if and only if [h(s) t[ =
0
. First observe that is well
dened. To see this suppose that s S and h H and there are distinct t and t
t
in
S such that [h(s) t[ = [h(s) t
t
[ =
0
. Then there exist i and j in s

bB
NC(h, b)
such that h(i) t and h(j) t
t
. But then, since i, j s
B
, there is some g in the
subgroup generated by B such that g(i) = j. Since i /

bB
NC(h, b) it follows that
h(j) = h(g(i)) = g(h(i)). Furthermore, g(h(i)) t because h(i) t and g belongs to
the subgroup generated by B. However, h(j) t
t
and t and t
t
are disjoint. Therefore
h(g(i)) ,= g(h(i)) contradicting the choice of i. A similar argument shows that is a
homomorphism.
Moreover, its image (H) is an abelian subgroup of S
S
. To see this, let s S. If
(g)((h)(s)) ,= (h)((g)(s)) then, g(h(s)) ,

h(g(s)) and hence there are innitely

many i s such that g(h(i)) ,= h(g(i)) contradicting that h almost commutes with g.
To begin, it will be shown that there cannot be a perfect set P S
S
such that:
(1) There is some g

H such that for all s S and P either (s) = s or

(s) = (g

)(s).
(2) Every element of P commutes with every element of (H).
To see this suppose that P and g

contradict the assertion. For P dene

S
by

(i) =
_
i if i s S and (s) = s
g

(i) if i s S and (s) ,= s.

It suces to show that

almost commutes with each h H. To see that this is so

let i N NC(g

, h) and let s S be such that i s. If (s) = s then h(

(i)) =
h(i) = (h(i)) the last equality holding because h(i) (h)(s) and ((h)(s)) =
(h)((s)) = (h)(s). On the other hand, if (s) ,= s then h(

(i)) = h(g

(i)) =
g

(h(i)) =

(h(i)) the last equality holding because h(i) (h)(s) and ((h)(s)) =
(h)((s)) ,= (h)(s).
It will now be shown that (H) is uncountable. Once this is done, since (H) has
already been shown to be abelian, it will follow from Lemma 3.3 that there exist P and
g

satisfying the conditions 1 and 2. So suppose that (H) is countable. To begin,

notice that there must be some A
(H)
such that h A
hH
is uncountable
keep in mind that A
B
. This so because if not, then it is easy to nd P and g

satisfying conditions (1) and (2). Simply let g

H be any permutation such that

(g

) A is dierent from the identity on innitely many sets in

(H)
. Then let P
MAXIMAL ABELIAN SUBGROUPS 11
be the set of all g S
S
such that for all A
(H)
either g A = (g

) A or else
g A is the identity. If no such g

exists then it follows that H is countable because

h A
hH
is countable for each A
(H)
and each h is the identity on all but
nitely many A
(H)
.
Hence, there must be some A
(H)
such that h A
hH
is uncountable and
hence there is some h

H such that h A : (h) A = (h

) A is uncountable.
Observe that if (h) A = (h
t
) A and there is some i A such that h(i) = h
t
(i)
then h A

h
t
A. To see this note rst that if i, j s A then there is
some b in the group generated by B such that b(i) = j. Since s / C it follows that
h(j) = h(b(i)) = b(h(i)) = b(h
t
(i)) = h
t
(j). If j A then there are s
i
and s
j
in
A such that i s
i
and j s
j
and there is

h H such that (

h)(s
i
) = s
j
. Since s
i
is innite, there is some i

s
i
(NC(

h, h) NC(

h, h
t
)) such that

h(i

) s
j
. Hence
h(

h(i

)) =

h(h(i

)) =

h(h
t
(i

)) = h
t
(

h(i

)) and, since

h(i

), j s
j
it follows that
h(j) = h
t
(j). But, since h A : (h) A = (h

) A is uncountable it is possible
to nd h and h
t
such that there are i and j in A such that h(i) = h
t
(i), h(j) ,= h
t
(j)
and (h) A = (h

) A. This is a contradiction.
The following alternate characterization, due to M. Bell, of the cardinal invariant p
of Denition 1.2 will be used in the proof of Theorem 3.2.
Theorem 3.1. The cardinal p is the least cardinal such that there is a -centred
partially ordered set
3
P and a collection D of p dense subsets of P for which there is
no centred subset G P intersecting each member of D.
Proof. See [3].
Theorem 3.2. If H S/F is an uncountable, maximal abelian subgroup then [H[ p
in other words, A([N]
<
0
) p.
Proof. If H S/F is an uncountable, maximal abelian subgroup and [H[ < p then it
follows from Lemma 3.5 that I

(o) : /F : o [H]
<
0
generates a proper
ideal.
Let P be the partial order consisting of all p = (h
p
, o
p
) such that:
(1) h
p
is a nite involution
4
(2) o
p
is a nite subset of S such that /F : o
p
H
and dene p q if and only if
(1) h
p
h
q
3
A partially ordered set is said to be -centred if it is the union of countably many centred subsets
in other words, it is the union of countably many subsets any nite subset of which has a lower
bound.
4
In other words, h
p
is it own inverse.
12 S. SHELAH AND J. STEPR

ANS
(2) o
p
o
q
(3) the domain of h
p
h
q
is disjoint from I

(o
q
)
(4) if j is in the domain of h
p
h
q
and o
q
then (j) is in the domain of h
p
h
q
and (h
p
(j)) = h
p
((j)).
It is clear that P is -centred because if h
p
= h
q
then the conditions p and q have
the common extension (h
p
, o
p
o
q
). Moreover, the sets D

= p P : o
p
are
dense for all S. Furthermore, so are the sets
E
n
= p P : n domain(h
p
) I

(o
p
).
To see that this is so, let p P be given and suppose that n / I

(o
p
). This, together
with Lemma 3.1, implies that the
S
p(n) is nite since all of the innite orbits under
o
p
are contained in I

(o
p
). Now let h
q
be the union of h
p
and the identity on
S
p(n)
and let q = (h
q
, o
p
). Then q E
n
and q p.
Hence, if [H[ < p then there is a lter G P meeting each D

for H and E
n
for n N. Dene
G
: N N by

G
(j) =
_
h
p
(j) if (p G) j domain(h
p
)
j if (p G)j / domain(h
p
)
.
It is easily veried that
G
S. To see that
G
/F commutes with each member of
H let /F H. Let p G be such that o
p
. Then if j N domain(h
p
) there
are two possibilities. If there is some q G such that j belongs to the domain of
h
q
it is clear that (
G
(j)) = (h
q
(j)) = h
q
((j)) =
G
((j)). However, if there is
no q G such that j belongs to the domain of h
q
then, by virtue of the fact that
E
j
G ,= , it must be the case that there is some q G such that j I

(o
q
). Since
o
q
it follows that (j) I

(o
q
). Hence
G
(j) = j and
G
((j)) = (j) and so
(
G
(j)) = (j) =
G
((j)).
All that remains to be shown is that the following sets are dense
D
,k
= p P : (j k) h
p
(j) ,= (k)
for H and k N because then a lter G P can be chosen meeting each relevant
D
,k
. To establish this, let p P be given. By Lemma 3.1 it follows that each element
of
S
p which is disjoint from I

(o
p
) is nite. Moreover, by Lemma 3.4 there must be
innitely many of the same cardinality, and, hence there must be distinct A and B in

S
p such that:
(1) both A and B are disjoint from I

(o
p
)
(2) both A and B are disjoint from the domain of h
p
(3) k < min(A)
(4) k < min(B)
(5) : A B is an o
p
-isomorphism.
MAXIMAL ABELIAN SUBGROUPS 13
There are two possibilities. If = A then let h
q
be the union of h
p
and the identity
on A and let q = (h
q
, o
p
). Otherwise, let h
q
= h
p

1
and let q = (h
q
, o
p
). In
either case q p and q D
,k
.
4. A([N]
<
0
) can be smaller than a
This section will establish that A([N]
<
0
) is smaller than a in the Laver model. The
argument will require some preliminary denitions and observations.
Denition 4.1. If T [N]
<
0
then T will be said to be small provided that:
If b a T then b T.
If ( is an innite subset of T then ( contains an innite -system.
A collection A of subsets of N will be said to be T-splitting if and only if for every
sequence (
n

n=0
each element of which is an innite set of pairwise disjoint subsets
of T there is X A such that [X]
<
0
(
n
,= for each n. Although the notion of
splitting will be sucient for most of the arguments to follow, at one point a more
complicated notion will be used. A collection A of subsets of N will be said to be
fully-T-splitting if and only if for every sequence ((
n
, Y
n
)

n=0
, such that each (
n
is
an innite set of pairwise disjoint subsets of T and Y
n
N, there is X A such that
for each n there is a (
n
such that
a
(X a) = Y
n
[a[; in other words, X is a copy
of Y
n
on the collapse of some member of (
n
.
The notion of splitting which already exists in the literature, for example in [5],
corresponds to what is here called fully-[N]
1
-splitting. It is worth noting that all the
small families considered here will be of bounded cardinality and, if all elements of
a small family T have cardinality less than k then in dening fully-T-splitting one
need not consider arbitrary sequences ((
n
, Y
n
)

n=0
but only those for which Y
n
is a
subset of k. In [5] A. Dow has shown that if A is splitting and W is obtained by
iterated Laver forcing over V then A is still splitting in W. Let L denote the Laver
partial order and let L

denote the countable support iteration of L of length . A

modication of the argument in [5] shows the following.
Theorem 4.1. If T is small and A is fully-T-splitting in V then
1
L

A is fully-

T-splitting
where is an ordinal.
Proof. The proof is almost the same as the proof of Lemma 9 on pages 245-247 of
[5]. One obvious change is that the A
n
of that proof are now required to enumerate
pairs ((
n
, Y
n
) such that (
n
is an L-name for an innite collection of pairwise disjoint
14 S. SHELAH AND J. STEPR

ANS
elements of T and Y
n
is an L-name for a subset of N. The only other change required
is that Fact 2 on page 246 must be replaced
5
by the following:
New Fact 2. If S L M and n there is S
t
S with the same root as S such
that the collection S
t
t) : S
t
t) M and S
t
t) (a (
n
)
a
(aX) = Y
n
[a[
is pre-dense below S
t
.
In order to prove this the following claim is required:
Claim. If T is a well founded tree such that each non-maximal node has innitely
many immediate successors and F is a function from the maximal nodes of T to T
then there is a subtree T
t
T such that
each maximal node of T
t
is a maximal node of T
each non-maximal node of T
t
has innitely many immediate successors
there is F
t
: T
t
T such that
if t t
t
are in T
t
then F
t
(t) F
t
(t
t
)
if t is a maximal element of T
t
then F
t
(t) = F(t)
if t

m T
t
and t

k T
t
and m ,= k then F
t
(t

m) F
t
(t

k) = F
t
(t).
Proof. The claim is easily proved by induction on the rank of T using the fact that T
is small.
In order to prove New Fact 2 let r be the root of S and let

be a name for the

value of the Laver real at [r[. Let T be the well founded tree whose maximal nodes
are minimal elements, t, of S such that there is some S
t
S such that the root of S
t
is t and
S
t

L


F
t
(
n
and min(

F
t
) >

and Y
n
max(

F
t
) = y
t

and dene F(t) = F

t
. Use the claim to nd a well founded tree T
t
T as well as a
function F
t
: T
t
T as in the conclusion of the claim. For any t T
t
which is not
maximal dene y
t
= y
t
[F
t
(t)[. Let T
tt
T
t
be a subtree such that r T
tt
and, if
s T
tt
is not maximal in T
tt
then s has innitely many immediate successors in t
tt
and the set of y
t

n
such that t

n T
tt
converges to z
t
N. Let
T

= t T
t
:
F

(t)
F
t
(t) X = z
t
[F
t
(t)[.
Observe that the requirement in the denition of F
t
that min(F
t
) >

guarantees that
F
t
(r) = and hence r T

. In fact, if t T

and t is not maximal in T then t has

innitely many successors in T

. To see this, it may as well be assumed that F

t
(t

m)
F
t
(t) ,= for all m such that t

m T
t
. It follows that F
t
(t

m) F
t
(t) : t

m T
t

is a pairwise disjoint, innite family in T and z

t
t
= n[F
t
(t)[
nz
t
N and both are in
5
The X in [5] seems to be a typographical error and should be changed to S. Also the [= there
is clearly intended to be .
MAXIMAL ABELIAN SUBGROUPS 15
M. Hence there are innitely many m such that
F

(t

m)\F

(t)
F
t
(t

m) F
t
(t) X =
z
t
t
[F
t
(t

m) F
t
(t)[ and, for any such m it follows that
F

(t

m)
F
t
(t

m) X =
z
t
[F
t
(t)[. In other words, t has innitely many successors in T

.
Let S
t
= T

S
t
: t max(T

). It is clear that, provided that S

t
L,
S
t

L
(a (
n
)
a
(a X) = Y
n
[a[. In order to see that S
t
L, let s S
t
. If
s t for some maximal t T

then s S
t
and, since t is the root of S
t
, it follows that
s has innitely many successors. Hence it may be assumed that s is a non-maximal
element of T

. It has already been established that s has innitely many successors

in T

and hence s has innitely many successors in S

t
.
The preceding result will not be used in full generality. For most of the argument
the following two corollaries will suce.
Corollary 4.1. If G L

is generic over V , B : N N is a function in V and

F V [G] is an innite function from D N to N such that F(d) < B(d) for d D
then there is a function H V such that H(d) = F(d) for innitely many d D.
Proof. Let T
B
be the set of singletons (n, m) : n N and m < B(n). Observe
that T
B
is small and that the set of functions in V bounded by B is T
B
-splitting in
V . Letting (
m
= (d, F(d)) : d D and d > m it follows that there is a function
H V such that H (
m
,= for each m. Hence H(d) = F(d) for innitely many
d D.
Corollary 4.2. If G L

is generic over V , B : N N is a function in V such that

B(n) > n for all n and F V [G] is an innite function from D N to N such that
B(d) < F(d) for d D then there is X N in V such that
for each x X there is no element of X between x and B(x)
and there are innitely many x X such that F(x) X.
Proof. Let T
B
be the set of pairs n, m : n N and m > B(n). Let A
B
be the
set of all innite subsets X N in V such that for each x X there is no element of
X between x and B(x). Observe that T
B
is small and that A
B
is T
B
-splitting in V .
Letting (
m
be an innite pairwise disjoint subset of d, F(d) : d D and d > m
it follows that there is X A
B
such that for each m there is a (
m
such that a X.
Hence X satises the corollary.
Lemma 4.1. If V satises 2

0
=
1
then there is an almost commuting family of
permutations T such that for each G L

1
which is a generic lter over V and for
each permutation h of N in V [G] V there is some T which does not almost
commute with h.
Proof. Let (q

, h

1
enumerate all pairs (q, h) such that
q
L

1
h is a permutation of N.
16 S. SHELAH AND J. STEPR

ANS
This enumeration will be used to construct involutions

1
by induction on . It
will be established that for each <
1
there is q q

such that either q

L

1
h

V
or there is such that
(4.1) q
L

1
[NC(h

)[ =
0
.
It is immediate that T =

1
will have the desired properties.
In order to describe the inductive construction, suppose that

have been
constructed. Let
i

i=0
enumerate and dene p
i
=

i
. Let
i
(k) =
p
n

i
n=0
(k)
and note that, because the

are almost commuting involutions it follows from

Lemma 3.1 that [
i
(k)[ 2
i
for all but nitely many k. Let
m
(k) denote the canonical
isomorphism type of
m
(k). To be more precise, let

m
(k) =
_
[
m
(k)[, p
1

1

m
(k)
, p
2

1

m
(k)
, . . . , p
m

1

m
(k)
,

m
(k)
(k),
_
and note the role of the constant determined by k. In particular, if n m and

m
(i) =
m
(k) then
n
(i) =
n
(k) because these can be dened from the constant using
the rst m permutations. Each
m
(k) will be referred to as an m-isomorphism type.
Next, dene

: N N such that for each j the following conditions hold:
(1) If is a j-isomorphism type and k N :
j
(k) = is nite then
j
(k) ,=
for each k

(j).
(2) If is a j-isomorphism type and k N :
j
(k) = is innite then there are

2
i=0
such that
(a)
j
(
i
) = for i < 3
(b)
j
(
n
)
j
(
m
) = if 0 n < m < 3
(c) j < min(
j
(
i
)) max(
j
(
i
)) <

(j) for i < 3.
(3) If x < j then max(
j
(x)) <

(j).
(4) If m and n are less than or equal to j then max(NC(p
n
, p
m
)) <

(j).
Finally, dene (j) =

(

(j))).
The induction depends on considering various cases.
Case 1. q

,
L

1
( < ) [NC(h

)[ <
0
.
In this case let q q

and < be such that q

L

1
[NC(h

)[ =
0
and note
that Condition 4.1 is satised. In this case simply let

be the identity permutation.

Case 2. Case 1 fails and q

,
L

1
h

/ V .
In this case simply let q q

and h be a permutation in V such that q

L

1
h

=

h.
Case 3. Case 1 and Case 2 both fail and there is some integer J such that
q

,
L

1
(k > J) h

J
(k) is a permutation of
J
(k).
MAXIMAL ABELIAN SUBGROUPS 17
Before continuing with this case, dene a partial function F by dening F(i) for
each integer i J to be the least integer such that there is some j such that
(4.2) i < j < F(i)
(4.3)
i
(j) =
i
(F(i))
(4.4)

(i) < min(
i
(j)) and

(i) < min(
i
(F(i)))
(4.5)
i
(j) ,=
i
(F(i))
(4.6) h

i
(F(i))
,= h

i
(j)
.
The rst thing to note is that q

1
(i N) F(i) is dened. In order to
see that this is so, suppose that q q

and i provide a counterexample; in other

words, q
L

F(

i) is not dened. It is possible to extend q to q

t
such that for each
i-isomorphism type there is a permutation h

such that
q
t

L

1
(k > i) i < min(
i
(k)) and
i
(k) = h

i
(k) = h

i
(k)
.
But this means that q
t
forces that h

is determined by h

: is an i-isomorphism type
and the value of h

on
i
(k) for those nitely many k such that

(i) ,< min(
i
(k)).
Hence, q
t

L

1
h

V contradicting that Case 2 fails.

Subcase 3A. q

1
(

n) F(n) > (n)

In this case if D = n N : (n) < F(n) then F D is an innite function. By
Corollary 4.2 it is possible to nd X V and q q

such that
(4.7) q
L

1
(

x X) F(x) X
and for each x X there is no element of X between x and (x). For x X let x

be
the least element of X greater than x. Choose L
x
< x

such that
x
(L
x
) =
x
(x

) and

(x) < min(

x
(L
x
)). To see that this is possible observe that x

> (x) >

(

(x))
and so k :
x
(k) =
x
(x

) must be innite by (1) in the denition of

. Hence L
x
can be found between

(x) and

(

(x)) using (2) in the denition of

. It follows that

x
(L
x
),
x
(x

)
commutes with p
m
if m x. Moreover, max(
x
(x

)) <

(x

). Hence,
if x and y are distinct members of X then the domain of

x
(L
x
),
x
(x

)
is disjoint
from

y
(L
y
),
y
(y

)
Hence, if

is dened by

(n) =
_

x
(L
x
),
x
(x

)
(n) if n
x
(L
x
)
x
(x

) for some x X
n otherwise
then

almost commutes with each p

m
and hence it almost commutes with each

.
It remains to show that q
L

1
[NC(h

)[ =
0
. To this end let x X be
such that F(x) also belongs to X. Let z be the greatest element of X below F(x);
in other words, F(x) = z

and so
z
(L
z
) =
z
(F(x)). Since x z it follows that
18 S. SHELAH AND J. STEPR

ANS

x
(L
z
) =
x
(F(x)) also. Furthermore, since F(x) is, by denition, the least integer
such that there some j such that Conditions 4.2, 4.3, 4.4, 4.5 and 4.6 hold with x in
the place of i it follows that
x
(j) =
x
(F(x)) =
x
(L
z
) and
h

x
(L
z
)
= h

x
(j)
for any such j. Since
h

x
(F(x))
,= h

x
(j)
it follows that

x
(L
z
),
x
(F(x))
does not commute with h

x
(L
z
). Because x z
this implies that

z
(L
z
),
z
(F(x))
does not commute with h

z
(L
z
). Since

z
(L
z
),
z
(F(x))

it must be that q
L

1
(n > x) h

(n)) ,=

(h

(n)). From Condition 4.7 it

follows that Condition 4.1 holds.
Subcase 3B. q

,
L

1
(

n) F(n) > (n)

Using Corollary 4.1 let q q

and H V be such that

q
L

1
(

n) H(n) = F(
n
(0)).
For each n choose, if possible, an integer L
n
such that

n
(0) < L
n
< H(n) <
n+1
(0)

(
n
(0)) < min(

n
(0)
(L
n
))

n
(0)
(L
n
) =

n
(0)
(H(n))
and then dene

by

(m) =
_

n
(0)
(L
n
),

n
(0)
(H(n))
(m) if m

n
(0)
(L
n
)

n
(0)
(H(n))
m otherwise.
Since

(
n
(0)) < min(

n
(0)
(L
n
)) and H(n) <
n+1
(0) for each n it follows, using (3)
in the denition of

, that the domains of

n
(0)
(L
n
),

n
(0)
(H(n))
and

m
(0)
(L
m
),

m
(0)
(H(m))
are disjoint if n ,= m; in other words, there is no contradiction in the denition of the
involution

. It is immediate from Condition 4.2 and 4.3 in the denition of F that if

H(n) = F(
n
(0)) then L
n
exists and, from Condition 4.6, that

n
(0)
(L
n
),

n
(0)
(H(n))
does not commute with h

. Hence Condition 4.1 holds.

Case 4. Neither of the cases 1, 2 or 3 holds.
In this case it may be assumed that
q

1
(i)(k)(j > k) h

i
(k) is not a permutation of
i
(k)
MAXIMAL ABELIAN SUBGROUPS 19
because otherwise Case 3 holds for some extension of q

. Since q

forces h

to be a
permutation it follows that
q

1
(i)(k)(j > k) h

(j) /
i
(j).
Let F
0
(i) be the least integer j > i satisfying
(4.8) h

(j) /
i
(j)
(4.9) (m i)(k min(
i
(j))) h

(p
m
(k)) = p
m
(h

(k))
(4.10) (

m)
i
(j) =
i
(m)
and dene F
1
(i) = h

(F
0
(i)).
Subcase 4A. F
1
(
n
(0)) <
n+1
(0) for all but nitely many n.
In this case, using Conditions 4.9 and 4.10 as well as (2) in the denition of

, it is
possible to conclude that for each n there are j
n
0
, j
n
1
and j
n
2
such that

n
(0)
(j
n
0
) =

n
(0)
(j
n
1
)
h

n
(0)
(j
n
1
) is a bijection from

n
(0)
(j
n
1
) onto

n
(0)
(j
n
2
)

n
(0)
(j
n
0
),

n
(0)
(j
n
1
) and

n
(0)
(j
n
2
) are all distinct

(
n
(0)) < min(

n
(0)
(j
n
i
)) for each i < 3.
Using Corollary 4.1, let : N N N N belonging to V and q q

be such that
(4.11) q
L

1
(

n) (n) = (
0
(n),
1
(n),
2
(n)) = (j
n
0
, j
n
1
, j
n
2
).
Note that, without loss of generality, it may be assumed that

n
(0)
(
0
(n)) =

n
(0)
(
1
(n))
for each n. As in previous cases, the domains of

n
(0)
(
0
(n)),

n
(0)
(
1
(n))
and

m
(0)
(
0
(m)),

m
(0)
(
1
(m))
are disjoint if n ,= m. Moreover, the domain of

n
(0)
(
0
(n)),

n
(0)
(
1
(n))
is disjoint
from

m
(0)
(
2
(m)) even in the case n = m. Hence, letting

be dened by

(m) =
_

n
(0)
(
0
(n)),

n
(0)
(
1
(n))
(m) if m

n
(0)
(
0
(n)),

n
(0)
(
1
(n))
m otherwise
it follows that

is an involution which almost commutes with each

for .
Furthermore,

is the identity on each

n
(0)
(
2
(n)).
Moreover, if (n) = (j
n
0
, j
n
1
, j
n
2
) then

is the identity on

n
(0)
(j
n
2
). Hence, in this
case,

(h

(h
1

(j
n
2
))) = j
n
2
. On the other hand, h
1

(j
n
2
) belongs to

n
(0)
(j
n
1
) and
so

(h
1

(j
n
2
)) belongs to

n
(0)
(j
n
0
). In particular, it does not belong to

n
(0)
(j
n
1
).
Since h

n
(0)
(j
n
1
) is a bijection from

n
(0)
(j
n
1
) onto

n
(0)
(j
n
2
) by Condition 4.9,
it follows that h

(h
1

(j
n
2
))) does not belong to

n
(0)
(j
n
2
) and, in particular, it is
20 S. SHELAH AND J. STEPR

ANS
not equal to j
n
2
. In other words, q forces that h

(h
1

(j
n
2
))) ,=

(h

(h
1

(j
n
2
))) and

n
(0) < h
1

(j
n
2
) <
n+1
(0). From Condition 4.11 it follows that Condition 4.1 holds.
Subcase 4B. There are innitely many n such that F
0
(
n
(0)) >
n+1
(0).
Using Corollary 4.2 nd q q

and X N such that X V , and

(4.12) q
L

1
(

x X) F
0
(x) X
and for each x X there is no element of X between x and (x). For each x X let
x

be the least element of X greater than x. For each such x there is some j
x
such
that
x < j
x
< x

(x) < min(
x
(j
x
))

x
(j
x
)
x
(x

) =

x
(j
x
) =
x
(x

)
Let

be dened by

(m) =
_

x
(j
x
),
x
(x

)
(m) if m
x
(j
x
)
x
(x

) for some x X
m otherwise.
Observe that if x X and F(x) = z

for some z X then q

L

1
h

(j
z
)
x
(j
z
)
since F
0
(x) = z

is dened to be the least integer k such that h

(k) /
x
(k). Hence
q
L

(h

(j
z
))
x
(z

). However,

(j
z
) = z

and so q
L

1
h

(j
z
)) =
h

(z

) /
x
(z

). It follows from Condition 4.12 that Condition 4.1 holds.

Subcase 4C. Neither Case 4A nor Case 4B holds.
Let T = [N]
2
and let A be the set of all innite subsets of N in V . Clearly, T is
small and A is fully-T-splitting. Let (
m
be a maximal pairwise disjoint set of pairs
i, j such that

i
(0) <
j
(0) < F
1
(
i
(0))
j+1
(0)
and i > m and let Y
m
= 0. Because Case 4A fails it follows that each (
m
is an
innite pairwise disjoint subset of T. Using Theorem 4.1 applied to ((
m
, Y
m
)

m=0
it is possible to nd X A such that there are innitely many x X such that

j
(0) < F
1
(
x
(0))
j+1
(0) and j / X.
Now let F(x) = F
0
(
x
(0)) for x X and note that F is also bounded by a function
in V because Case 4B fails. Using Corollary 4.1 it is possible to nd q q

and f in
V such that
(4.13) q
L

1
(

x X) f(x) = F(x).
For each x X choose, if possible, j
x
such that

x
(0) < j
x
<
x+1
(0)
MAXIMAL ABELIAN SUBGROUPS 21

(
x
(0)) < min(
x
(j
x
))

x
(0)
(j
x
) =

x
(0)
(f(x))

x
(0)
(j
x
) ,=

x
(0)
(f(x)).
and observe that if f(x) = F(x) then it is possible to nd such a j
x
. This is so
because, by the denition of F
0
, there are innitely many j such that

x
(0)
(j) =

x
(0)
(F
0
(
x
(0))) =

x
(0)
(f(x)). As in previous cases, let

be dened by

(m) =
_

x
(0)
(j
x
)),

x
(0)
(f(x))
(m) if m

x
(0)
(j
x
))

x
(0)
(f(x)) for some x X
m otherwise.
Just as in previous cases, it is easy to check that

is well dened and almost commutes

with each

for . Moreover,
(4.14) (j / X)(m /

j1
(0)
(f(j 1)))
j
(0) < m
j+1
(0)

(m) = m.
Now suppose that x X and f(x) = F(x) and

j
(0) < F
1
(
x
(0))
j+1
(0)
and j / X. Then

(j
x
) = f(x) = F
0
(
x
(0)) and hence h

(f(x)) = h

(F
0
(
x
(0))) =
F
1
(
x
(0)) and so F
1
(
x
(0)) /

x
(0)
(f(x)). From Observation 4.14 it may be concluded
that

(F
1
(
x
(0))) = F
1
(
x
(0)). Next, note that since j
x
,= f(x) it follows that
h

(j
x
) ,= h

(f(x)) = F
1
(
x
(0)) and so

(h

(j
x
)) ,=

(F
1
(
x
(0)) = F
1
(
x
(0)) =
h

(j
x
)). From Condition 4.13 it follows that Condition 4.1 holds.
The following result, due to S. Shelah, is 5.31 in [13]. It will suce to know that
Laver forcing L is NEP without having to dene the concept.
Lemma 4.2. Let B

1
be a family of Borel sets in a model of set theory V and
suppose that V [=

1
B

= . Let P be a NEP partial order with denition in V

and suppose that P

2
is a countable support iteration such that P
+1
= P

P for
every
2
. If
1
P

1
B

=
then
1
P

1
B

= .
Theorem 4.2. It is consistent that A([N]
<
0
) =
1
< a.
Proof. The model witnessing this is the one obtained by forcing with L

2
over a model
V satisfying 2

0
=
1
. From Lemma 4.1 it follows that there is an almost commuting
family of permutations T such that for each G L

1
which is a generic lter over V
and for each permutation h of N in V [G]V there is some T which does not almost
commute with h. Then if Q is any maximal almost commuting family of permutations
22 S. SHELAH AND J. STEPR

ANS
in V containing T it follows that if G L

1
which is a generic lter over V then for
each permutation h of N in V [G] there is some Q such that either = h or
does not almost commute with h. In other words, letting B

be the Borel set of all

permutations of the integers which almost commute with but are not equal to
1
L

= .
It follows from Lemma 4.2 that
1
L

=
or, in other words, 1
L

2
A([N]
<
0
) =
1
. The fact that a =
2
in this model is
well known and can be found, for example, in [2].
5. The Cohen model and the summable ideals
The remaining results will deal with quotients of groups of permutations of the
integers with respect to ideals other than the ideal of nite sets. Since Section 6 will
be devoted to establishing that a(1) < A(1) = 2

0
for a certain ideal, it is natural to
ask whether the phenomenon A(1) < 2

0
might not be a peculiarity of the nite ideal.
This section will show that this is not the case. The ideal 1
1/x
is dened to be the
set of all X N such that

xX
1/x < . It will be shown that S(1
1/x
)/F(1
1/x
) has
a maximal abelian subgroup of size
1
in any model obtained by adding uncountably
many Cohen reals. The basic scheme of the argument is that in a model of the
form V [c

1
] where c

1
are Cohen reals, it is possible to dene permutations

1
which almost commute such that

V [c

+1
] and such that they are
close to maximal in the following sense: Given any permutation which is not rst
order denable using elements of

1
as parameters, there is some
1
such
that NC(,

) / 1
1/x
. In other words, if G is the group generated by

1
, and
G
t
G/F(1
1/x
) is any maximal abelian group, each element of which is the equivalence
class of some permutation which is rst order denable using elements of

1
as
parameters, then the cardinality G
t
is the same as that of G and, moreover, G
t
is a
maximal abelian subgroup of S(1
1/x
)/F(1
1/x
). The rest of the section will concentrate
on the construction of

1
. Consequently, many of the results of this section will
assume as a hypothesis a family of permutations with certain properties. These can
protably be thought of as the permutations obtained from the Cohen reals at some
stage of the transnite induction.
Denition 5.1. A family of permutations T will be said to be tame if
(1) each T is an involution
(2) each pair of permutations in T almost commute modulo the ideal 1
1/x
MAXIMAL ABELIAN SUBGROUPS 23
(3) if a T is nite then
a
consists of nite sets and
F(a) =
_
x
a
: ( a)(n x) (n) = n
belongs to the ideal 1
1/x
and

a
will denote x
a
: x , F(a)
(4) for every nite a T there is (a) [

a
]
<
0
and a family
a
x,y

x,y

a
\(a)
such that
(a) each mapping
a
x,y
is an a-isomorphism from x to y
(b)
a
y,z

a
x,y
=
a
x,z
(c) if a b then, except for a nite set of exceptions, if x
0
and x
1
belong to

b
then for all y
0

a
such that y
0
x
0
there is some y
1

a
such that

b
x
0
,x
1
y
0
=
a
y
0
,y
1
.
The notation
a
x,y
will be used to represent
a
x,y

a
y,x
.
The following lemma provides a method for enlarging tame families. This will be
used in the transnite induction mentioned in the introduction to this section.
Lemma 5.1. If T is a tame family of permutations and this is witnessed by

a
x,y
: a [T]
<
0
and x, y

a
(a)
and a
n

n=0
is a family of nite subsets of T such that
a
n
a
n+1
for each n

n=0
a
n
= T
x
n
, y
n
[

a
n
(a
n
)]
2
for each n
(x
n
y
n
) (x
m
y
m
) = unless n = m
N

n=0
(x
n
y
n
) 1
1/x
for each nite subset a T and for all but nitely many n, if x

a
(a)
and x x
n
then there is some x
t

a
(a) such that
a
n
x
n
,y
n
x =
a
x,x

and =

n=0

a
n
x
n
,y
n

then T is also tame.

Proof. The fact that almost commutes with each member of T is immediate from
the fact that each
a
x,y
is an a-isomorphism. Since each
a
x,y
is an involution, so
is . This also yields that
a
consists of nite sets. For any nite a T it is
immediate that
F(a ) F(a)
_
N

_
n=m
x
n
y
n
_
1
1/x
where m is chosen large enough that a a
m
.
The nite sets (a)

a
for a [T ]
<
0
and the family
_

a
x,y
: a [T ]
<
0
and x, y

a
(a)
_
witnessing that T is tame must be dened. If a T dene (a) = (a) and

a
x,y
=
a
x,y
for x, y

a
(a). In order to dene (a ), start by using the
24 S. SHELAH AND J. STEPR

ANS
hypothesis of the lemma to nd an integer

(a ) so great that if n

(a )
and x

a
(a) and x x
n
then there is some x
t

a
such that
a
n
x
n
,y
n
x =
a
x,x
.
Then dene
(a ) = (a)
_
i<

(a)
x

a
: x x
i
y
i

and note that it is nite since the x

i
are nite. Observe that if z

a
(a
) then z x
m
y
m
for some m

(a ). Denote this integer m by m(z).
Consequently, if z

a
(a ) then z = z
x
z
y
where z
x
x
m(z)
and
z
y
y
m(z)
and both z
x
and z
y
belong to

a
. Given z and z in

a
(a )
dene

a
z,w
=
a
z
x
,w
x
a
z
y
,w
y
and note that
a
z,w
is an a -isomorphism. It is also routine to verify that
Condition 4b holds. To see that Condition 4c is satised let a b T be nite.
Let a
t
= a and b
t
= b . It may as well be assumed that b since otherwise
the hypothesis that T is tame can be applied directly. Let z
0
and z
1
in

b
(b) be
arbitrary and let w
0

a
be such that w
0
z
0
. Then

b
z
0
,z
1
w
0
=
_

z
x
0
,z
x
1

b

z
y
0
,z
y
1
_
w
0
=
b

z
x
0
,z
x
1
(w
0
z
x
0
)
b

z
y
0
,z
y
1
(w
0
z
y
0
)
=
a

w
0
z
x
0
,w
x
1

a

w
0
z
y
0
,w
y
1
for some w
x
1
and w
y
1
provided that neither w
0
z
x
0
nor w
0
z
y
0
come from the nite
set of exceptions to Condition 4c for a
t
and b
t
. There are then two cases to consider.
If a
t
= a then, because w
0

a
, it must be that either w
0
z
x
0
= or w
0
z
y
0
= .
Hence either

w
0
z
x
0
,w
x
1

a

w
0
z
y
0
,w
y
1
=
a
w
0
,w
x
1
or

w
0
z
x
0
,w
x
1

a

w
0
z
y
0
,w
y
1
=
a
w
0
,w
y
1
and in either case the result is established. On the other hand, if a ,= a
t
then a
and since w
0

a
it follows that (w
0
z
x
0
) = w
0
z
y
0
. Hence (w
x
1
) = w
y
1
and so,
letting w
1
= w
x
1
w
y
1
,

w
0
z
x
0
,w
x
1

a

w
0
z
y
0
,w
y
1
=
a
w
0
,w
1
as required.
For the rest of this section some simplifying notation will be introduced to refer to
closed sets of orbits associated with families of permutations. The elements of

a
will
be enumerated as
i
a

i=0
in such a way that i < j if and only if min(
i
a
) < min(
j
a
).
If T is a tame family and a is a nite subset of T then, in this context, (a) will be
an integer such that Condition 4 in Denition 5.1 holds for all
i
a
with i (a); in
other words, the nite set of exceptions to Condition 4 is contained in
i
a

(a)
i=0
. The
MAXIMAL ABELIAN SUBGROUPS 25
following denition describes a partial order which can be used to create a permutation
satisfying the hypothesis of Lemma 5.1.
Denition 5.2. Given a tame family T such that this is witnessed by

a
i,j
: a [T]
<
0
and i, j (a)
and such that it also satises
(5.1) ( T) lim
n
(n)
n
= 1
dene the partial order P(T) to consist of all pairs
p =
_
(a
p
m
, i
p
m
, j
p
m
)
k
p
m=0
,
p
_
=
_
(a
m
, i
m
, j
m
)
k
m=0
,
_
such that, letting D(p) =

k
m=0

i
m
a
m

j
m
a
m
,
(1) k N
(2) each a
m
is a nite subset of T
(3) a
m
a
m+1
(4) i
m
and j
m
are distinct integers greater than (a
m
)
(5) (
i
m
a
m

j
m
a
m
) (
i
n
a
n

j
n
a
n
) = unless n = m
(6) if a a
k
and
i, j > j
k

i

a

i
a
k
then
a
k
i,j

i

a
=
a
i

,j
for some j
t
(7) > 0
and if
= sup
uN\D(p)
_
sup
a
k

1
(u)
u

_
then
(5.2)
_
1 + 2

_
1
1
(1 + )

+
j
k
+ 1
min(
k
a
k
)
__
(1 + )

< 1 +
where = [a
k
p[. Also,
(5.3)

n<j
k
and n/ D(p)
1/n +

n>j
k
and yF(a)
1/n < 1.
Dene p q if
p

q
and (a
p
m
, i
p
m
, j
p
m
) = (a
q
m
, i
q
m
, j
q
m
) for m k
q
and
(5.4) 1 <

a
p
m
i
p
m
,j
p
m
(u)
u
< 1 +
for m > k
q
and u
i
p
m
a
p
m
.
The following technical lemma will be useful in applying the partial order P(T).
26 S. SHELAH AND J. STEPR

ANS
Lemma 5.2. Suppose that T is a tame family and a is a nite subset of T and = [a[.
Suppose also that > 0 and m N. and, furthermore that
(5.5) (i m)(x
i
a
)( a)

1
(x)
x

<
(5.6)

kF(a)
1/k < log
_
1 + 2
1 +
_
.
Then the following inequalities hold for any i m and any k:
(5.7)
max(
i
a
)
min(
i
a
)
< (1 + )

(5.8)
min(
i+1
a
)
min(
i
a
)
< 1 + +
m2

min(
i
a
)
(5.9)
min(
i+k
a
)
min(
i
a
)
<
_
1 + +
m2

min(
i
a
)
_
k
(5.10) (1 )

<

a
i,i+k
(n)
n
<
_
1 + +
m2

min(
i
a
)
_
k
(1 + )

Proof. In order to prove 5.7 the rst thing to note is that if x, y

i
a
then there is
k and a sequence (
1
,
2
, . . . ,
k
) a
k
such that
x =
1

2
. . .
k
(y).
Given x
i
a
let k(x) be the least integer such that
x =
1

2
. . .
k(x)
(min(
i
a
))
and proceed by induction on k(x) to show that
x
min(
i
a
)
< (1 + )

for every x
i
a
. If k(x) = 0 then x = min(
i
a
) and the result is clear. Suppose
that the lemma has been established for all x such that k(x) = n. Given x such that
k(x) = n+1 it is possible to nd x
t
such that k(x
t
) = n and x = (x
t
) for some a.
From 5.5 it follows that
(x
t
)
x
t
< 1 +
and from the induction hypothesis it follows that
x
t
min(
i
a
)
< (1 + )
n
.
and, hence,
x
min(
i
a
)
< (1 + )
n+1
MAXIMAL ABELIAN SUBGROUPS 27
as desired.
To see that 5.8 holds begin by observing that if m i
t
i and
i

a
min(
i
a
) ,=
then, by 5.7,
min(
i
a
) < max(
i

a
) < min(
i

a
)(1 + )

and hence
min(
i

a
) >
min(
i
a
)
(1 + )

.
Therefore, the cardinality of
_
mi

a
min(
i
a
)
is no greater than
_
min(
i
a
)
min(
i
a
)
(1 + )

_
2

= min(
i
a
).
Using that
min(
i
a
)+min(
i
a
)2

n=min(
i
a
)+min(
i
a
)
1
n
< log
_
1 + 2
1 +
_
it follows that F(a)

mi

i

i

a
does not cover the interval of integers between min(
i
a
)
and min(
i
a
) + min(
i
a
)2. Hence
min(
i+1
a
) < min(
i
a
) + min(
i
a
)2 + m2

as required.
The general statement 5.9 follows by repeated application of 5.8.
To prove 5.10 let n
i
a
. Combining 5.7 and 5.9 yields

a
i,i+k
(n)
n

max(
i+k
a
)
min(
i
a
)
<
_
1 + +
m
min(
i
a
)
_
k
(1 + )

establishing the last half of the inequality. For the rst half note that min(
i
a
)
min(
i+k
a
) and hence, from 5.7 and 5.9 it follows that

a
i,i+k
(n)
n

min(
i+k
a
)
max(
i
a
)

min(
i
a
)
max(
i
a
)
>
1
(1 + )

> (1 )

Lemma 5.3. If p P(T) and j j

p
k
p
then there is q p such that
j
a
D(q).
Proof. It suces to prove this for the case that j = 1 + j
p
k
p
. Let
q = ((a
q
m
, i
q
m
, j
q
m
)
k
p
+1
m=0
,
p
)
where (a
q
m
, i
q
m
, j
q
m
) = (a
p
m
, i
p
m
, j
p
m
) if m k
p
and i
q
k
p
+1
= j, j
q
k
p
+1
= j + 1 and a
q
k
p
+1
=
a
k
p. From Condition 5.2 of Denition 5.2 and Conclusion 5.10 of Lemma 5.2, with
k = 1 it follows that Requirement 5.4 is satised and so q P(T) and q p.
28 S. SHELAH AND J. STEPR

ANS
Lemma 5.4. If p P(T) and > 0 and a N is nite then there is q p such that

q
and a
q
k
q
a.
Proof. Let b = a a
p
k
p
. First apply Lemma 5.3 to extend p to q
t
so that the do-
main of D(q
t
) is suciently large that Condition 6 of Denition 5.2 holds as well as
Condition 5.2 with
= sup
uN\D(q

)
_
sup
b

1
(u)
u

_
and
p
replaced by . It can also be arranged that (b) < max(D(q
t
)) and that

n<j
k
and n/ D(q

)
1/n +

n>j
k
and yF(b)
1/n < 1.
Then let be the least integer such that
j
b
D(q
t
) = and let
q =
_
(a
q
m
, i
q
m
, j
q
m
)
k
q

+1
m=0
,
_
where (a
q
m
, i
q
m
, j
q
m
) = (a
q

m
, i
q

m
, j
q

m
) if m k
q

and i
q
k
q

+1
= , j
q
k
q

+1
= + 1 and a
q
k
q

+1
=
b. Note that just as in the proof of Lemma 5.3 it follows that q P(T) and q q
t

p.
Corollary 5.1. If G P(T) is generic dene (a
n
, i
n
, j
n
) = (a
p
n
, i
p
n
, j
p
n
) for some (or,
equivalently, any) p G. If
G
is dened

G
=

_
n=0

a
n
i
n
,j
n

then
G
almost commutes with each member of T and the family T
G
is tame.
Proof. From Denition 5.2 it follows that
G
is an involution and from Lemma 5.4
that it almost commutes with each member of T. From Lemmas 5.3 and 5.4 it follows
that

n=0
a
n
= T and N

n=0
(i
n
j
n
) 1
1/x
. It follows from Lemma 5.1 that the
family T
G
is tame.
Lemma 5.5. If p P(T) and S(1
1/x
) but is not rst order denable using
nitely many parameters from T and k N then there is q p such that
q
P(T)

i=k
1/i :
G
((i)) ,= (
G
(i)) > 1
where
G
is as dened in Corollary 5.1.
Proof. As a convenience, let a = a
p
k
p
, = [a[. Let t

be such that Hypothesis 5.6 of

Lemma 5.2 holds with m = t

and then choose t t

to be some integer such that the

inequalities
1
p
<
(x)
x
< 1 +
p
MAXIMAL ABELIAN SUBGROUPS 29
hold for any j > t and x
j
a
. By appealing to Lemma 5.3 it may assumed that
t k
p
. A nal application of Lemma 5.3 will allow the assumption that if
= sup
jN\D(p)
_
sup
a

1
(j)
j

_
then
(5.11)
_
1 + 2

_
1
1
(1 + )

+
t + 1
min(
J
a
)
__
6
(1 + )

< 1 +
p
.
where J = k
p
+ 1.
For use later on, let be so small that
2
1 2

6
< 1 +
p
and choose small enough that
(5.12)
2 (1 + 2

((1 + )

1)) (1 + )

1 2

6( + (1 + )

1)
< 1 +
p
.
Using Lemma 5.3, is may be assumed
6
that
1 <
(x)
x
< 1 +
for all x / D(p) and a. A nal application of Lemma 5.3 yields that it may also
be assumed that
(5.13)

nF(a)\t
1/n < log(2).
The following fact will play a role later in the proof but is included here to explain
the signicance of the exponent 6 in Inequality 5.11 as well as in the indexing to follow.
Claim 4. Given any Sym(6) other than the identity there is Sym(6) without
xed points such that is an involution and does not commute with .
Proof. The proof is elementary.
Dene E
i
=

5
w=0

J+6i+w
a
for each i N. Given an involution without xed points
H of some interval of integers [J, J + 2K] let p
H
= ((a
m
, i
m
, j
m
)
k
p
+K
m=0
,
p
) where
(a
m
, i
m
, j
m
) =
_
(a
p
m
, i
p
m
, j
p
m
) if m k
p
(a, i
m
, H(i
m
)) if m > k
p
and i
m

k
p
+K
m=k
p
+1
enumerates a maximal subset of the domain of H which is disjoint
from its image under H. Note that it may turn out to be that case that p
H
/ P(T)
because it is possible that, for example,

a
i
m
,H(i
m
)
(n)
n
> 1 +
p
6
Actually, it must also be observed that Lemma 5.3 does not require changing the value of
p
.
30 S. SHELAH AND J. STEPR

ANS
for some n
i
m
a
. However,
(5.14) if [i
m
H(i
m
)[ < 6 then 1
p
<

a
i
m
,H(i
m
)
(n)
n
< 1 +
p
by Condition 5.11 and Conclusion 5.10 of Lemma 5.2.
Observe that if X N then by Conclusion 5.9 of Lemma 5.2,
(5.15) X 1
1/x
if and only if E(X) 1
1/x
where E(X) =

XE
j
,=
E
j
. This will be used repeatedly in order to restrict the
possible structure of .
To begin, let W : N N be dened so that if x
i
a
then (x)
W(x)
a
. First
note that if there exist x
1
and x
2
in
u
a
such that W(x
1
) ,= W(x
2
) then, because
a
is a minimal set closed under the permutations in a, there must be some permutation
in the group generated by a such that (x
1
) = x
2
and, hence, ((x
1
)) ,= ((x
1
)).
Therefore, by 5.15, it may be assumed that if Z is the set of all z such that there is
y
a
(z) such that W(z) ,= W(y) then E(Z) 1
1/x
. Let

zZ
1/z = s
Z
.
Next, let W
t
be dened for x / E(Z) such that if x E
i
then
W(x)
a
E
W

(x)
. Let
X = x N : (i N)(y E
i
) x E
i
and W
t
(y) ,= W
t
(x)
and suppose that E(X) / 1
1/x
. For each i such that E
i
Z = and E
i
X ,=
choose y
i
E
i
and x
i
E
i
such that W
t
(y
i
) ,= W
t
(x
i
) and let
i
be a xed point free
involution of J + 6i + n
5
n=0
such that if y
i

w
y
a
and x
i

w
x
a
then
i
(w
y
) = w
x
.
(Note that w
y
,= w
x
since E
i
Z = .). Then, using 5.15, it is possible to choose
K N such that

K
i=0
1/y
i
> 1 +s
Z
. If E
i
Z ,= and 0 i K let
i
be any xed
point free involution of J +6i+n
5
n=0
. Let H =

k
i=0

i
and note that q = p
H
P(T)
by Observation 5.14 and q p. It follows from the choice of K and the denition of
X that q satises the requirements of the lemma.
Hence, it may be assumed that E(X) 1
1/x
and W
t
is constant on E
i
provided that
E
i
NE(X). Let Y = i N : E
i
NE(X). Let W
tt
be dened on Y such that
if x E
i
then
W

(x)
a
E
W

(i)
. Therefore there is a partition Y = Y
0
Y
1
Y
2
Y
3
such that W
tt
(Y
i
) Y
i
= for each i 3 and W
tt
is the identity on Y
3
. Let j 4 be
such that

iY
j
E
i
/ 1
1/x
. Observe that for each i Y
j
there is a permutation
i
of 6
satisfying that if z
J+6i+u
a
then W(z) = J + 6W
tt
(i) +
i
(u).
First assume that j 3. Choose
i
and
i
to be any involutions of 6 without
xed points such that
i
(
i
(u)) ,=
i
(
i
(u)) for some u < 6. It follows that if z

J+6i+u
a
then
a
J+6i+u,J+6i+
i
(u)
(z)
J+6i+
i
(u)
a
. Moreover, (z)
J+6W

(i)+
i
(u)
a
and (
a
J+6i+u,J+6i+
i
(u)
(z))
J+6W

(i)+
i
(
i
(u))
a
. However,

a
J+6W

(i)+
i
(u),J+6W

(i)+
i
(
i
(u))
((z))
J+6W

(i)+
i
(
i
(u))
a
,=
J+6W

(i)+
i
(
i
(u))
a
.
MAXIMAL ABELIAN SUBGROUPS 31
Therefore if q forces
G
to contain both

a
J+6i+u,J+6i+
i
(u)
and
a
J+6W

(i)+
i
(u),J+6W

(i)+
i
(
i
(u))
this will guarantee that
G
((z)) ,= (
G
(z)) for z
J+6i+u
a
.
With this in mind, ket K be such that

iKY
j
1
max(E
i
)
> 1
and let
H =
_
iKY
j

i

W

(i)
and note that p
H
p by Observation 5.14. The choice of K guarantees that q = p
H
is as required by the lemma.
Hence, assume that j = 3. If the set of i Y
3
such that
i
is not the identity is not
in 1
1/x
then using Claim 4 it is possible to choose, for each i Y
3
, an involution
i
of 6 without xed points which does not commute with
i
. Using an argument very
similar to the previous case it is possible to nd a suciently large K so that if
H =
_
iKY
3

i
then p
H
is as required by the lemma. Notice that since there are no
i
in this case,
Claim 4 must be used in this argument.
Therefore, by omitting a set in 1
1/x
, it may be assumed that
i
is the identity for
all i Y
3
. For any permutation of
J
a
and z 6 let Y (, z) be the set of all i Y
3
such that

J+6i+z,J

J,J+6i+z

J
a
= .
If for each z 6 there is only one permutation
z
of
J
a
such that

iY (
z
,z)

J+6i+z
a
/
1
1/x
then /F(1
1/x
) can be dened from a and
z

z6
. So it may be assumed that
it is possible to choose z 6 and a permutation
z
of
J
a
such that if U
0
= Y (
z
, z)
and U
1
= N U
0
then

iU
0

J+6i+z
a
/ 1
1/x
and

iU
1

J+6i+z
a
/ 1
1/x
. For j U
1
let

j
=
J+6j+z,J

J,J+6+z
and note that
j
,= . The key point to keep in mind is
that if i U
0
and j U
1
then

a
J+6j+z,J

_

a
J+6i+z,J+6j+z

1
(
a
J+6i+z,J+6j+z
)
1
_

a
J,J+6j+z
=
_

a
J+6j+z,J

a
J,J+6j+z
_

a
J+6i+z,J

1

a
J,J+6i+z
_
=
1
j
and it follows that
a
J+6i+z,J+6j+z

1
(
a
J+6i+z,J+6j+z
)
1
is dierent from the
identity. Therefore if H is any involution such that H(J +6i+z) = J +6j +z for i U
0
and j U
1
then then there is x E
i
such that the inequality
G
((x)) ,= ((x)) is
forced by p
H
. However, notice that, unlike all previous cases, i and j are not equal and
so, if [i j[ is too large then Requirement 5.4 may fail for p
H
. The remainder of the
32 S. SHELAH AND J. STEPR

ANS
argument is devoted to showing that there are suciently many pairs (i, j) U
0
U
1
such that p
H
satises Requirement 5.4 as well as the conclusion of the lemma.
To this end, let U = n U
0
: n+1 U
1
. For n U let n
0
be the greatest integer
such that the interval [n n
0
, n] is contained in U
0
and let n
1
be the largest integer
such that [n+1, n+1+n
1
] U
1
. Let U

0
be the set of all n U such that n
0
n
1
and
U

1
be the set of all n U such that n
0
> n
1
and dene U
t
i
=

nU

i
[nn
0
, n+n
1
+1]
and observe that Y
3
= U
t
0
U
t
1
. Hence, either U
t
0
or U
t
1
fails to belong to 1
1/x
. In either
case the following argument is similar so assume that U
t
0
/ 1
1/x
.
Recall the denitions of and at the beginning of the proof. It will rst be shown
that if
m > n > m(1 )
then
(5.16) 1
p
<

a
J+6n+u,J+6m+u
(i)
i
< 1 +
p
for any i
J+6n+u
a
. Keep in mind that min(E
i
) = min(
J+6i
a
) for any i. Begin
by observing, using Conclusion 5.7 of Lemma 5.2, that if
j
a
intersects the interval
[min(E
n
)(1 + )

, min(E
m
)] then J + 6n j J + 6m. Hence,
(5.17) N [min(E
n
)(1 + )

, min(E
m
)]
m
_
z=n
E
z
F(a).
Note that
[F(a) [min(E
n
)(1 + )

, min(E
m
)][ min(E
m
)/2
because of Condition 5.13. Therefore,
(5.18) min(E
m
) min(E
n
)(1 + )

6(mn) + min(E
m
)/2.
It follows that
min(E
m
)/2 min(E
n
) 2

6(mn) + (min(E
n
)(1 + )

min(E
n
))
2

6(mn) + 2

(J + 6n)((1 + )

1)
2

(6(mn) + 6n((1 + )

1)) + 2

J((1 + )

1)
2

(6(mn) + 6m((1 + )

1)) + 2

J((1 + )

1)
2

6m( + (1 + )

1) + 2

J((1 + )

1)
and hence,
min(E
m
)
2 min(E
n
)
1
m
min(E
n
)
(2

6( + (1 + )

1)) + 2

J((1 + )

1).
Therefore, using the fact that m min(E
m
) and J min(E
n
)
min(E
m
)
min(E
n
)

2 (1 + 2

((1 + )

1))
1 2

6( + (1 + )

1)

1 +
p
(1 + )

MAXIMAL ABELIAN SUBGROUPS 33

by Inequality 5.12. Therefore, using Conclusion 5.7 of Lemma 5.2

a
J+6n+u,J+6m+u
(i)
i

max(E
m
)
min E
n
1 +
p
for any u < 6 and i E
n
. Similar reasoning shows that both Inequalities 5.16 hold.
Consequently, dening H so that H(J + 6n + u) = J + 6m + u will not conict with
Condition 5.4 holding for p
H
.
The only question which remains is whether it is possible to add enough of these
extensions to provide a large witness to
G
not commuting with . In case there is
some K such that n
0
n(1 ) for all n K it follows that

nU

0
\K

n
j=n
0
1/j = .
Moreover, for each n K and j such that n
0
j n there is some x
j
E
j
such that
(5.19) (
a
J+6j+z,J+6(j+n
0
)+z
(x
j
)) ,=
a
J+6j+z,J+6(j+n
0
)+z
((x
j
))
since j + n
0
< n + n
1
. Hence, by 5.15, it follows that

jU

0
\K
1/x
j
= and so it is
possible to choose M so that

M
jU

0
\K
1/x
j
1. Dening H so that H(J + 6j +z) =
J+6(j+n
0
)+z for n U
t
0
such that K n M and n
0
j n will satisfy the lemma
because, in this case, n(1 ) n
0
j n and so p
H
P(T) by Observation 5.16.
In the other case, there is an innite set U
tt
U
t
0
such that n
0
< n(1 ) for each
n U
tt
. It follows that if n U
tt
and n(1 ) j n then there is some x
j
E
j
such that
(
a
J+6j+z,J+6(j+n(1 )|)+z
(x
j
)) ,=
a
J+6j+z,J+6(j+n(1 )|)+z
((x
j
))
holds. Using Lemma 5.2 it follows that for n U
tt
n

i=n(1 )|
1/x
i

n

i=n(1 )|
1
max(E
i
)

n

i=n(1 )|
1
min (
J+6i+5
a
) (1 + )
a

i=n(1 )|
1
2

(J + 6i + 5)2(1 + )
a
and elementary calculations, using Condition 5.13 show that the limit as n increases
to innity of the last term of the inequality is
1
2

12(1 + )
a
ln
_
1
1
_
= > 0.
Now it suces to choose a nite subset T U
tt
such that
[T[ > 2/
and
n

i=n(1 )|
1
2

(J + 6i + 5)2(1 + )
a
>

2
34 S. SHELAH AND J. STEPR

ANS
for all n T. Then dene H(J + 6i + z) = J + 6(i + n(1 )|) + z for n T and
n(1 ) i n and note that setting q = p
H
p as before satises the requirements
of the lemma.
Theorem 5.1. It is consistent that A(1
1/x
) =
1
< 2

0
.
Proof. Let V be a model where 2

0
>
1
and let V
t
be obtained from V by adding
1
Cohen reals. To be precise, V
t
=

1
V

where

and V
+1
= V

[G

]
where G

is Cohen generic over V

for the partial order P(

). Moreover,

. Using Lemmas 5.3 and 5.4 and Lemma 5.1 it follows that

is a tame
family whose elements almost commute for each
1
. Let

1
be a
maximal almost abelian subgroup of the subgroup of all S(1
1/x
)/F(1
1/x
) which
are rst order denable from some nite subset of

1
. To see that is maximal
in S(1
1/x
)/F(1
1/x
) suppose that V [

]. If is rst order denable from some

nite subset of

then either or there is some such that NC(, ) /

1
1/x
. On the other hand, if is not rst order denable from some nite subset of

then by Lemma 5.5 and genericity it follows that NC(,

) / 1
1/x
.
6. It is possible that a(1) < A(1)
Since it has been shown in Proposition 2.1 that A([N]
<
0
) a it is natural to wonder
whether there might not be a more general result asserting that A(1) is bounded by
a(1) as dened in Denition 1.2. It will be shown that no such result holds, at least
not in the generality indicated.
Fix an increasing sequence of integers ^ = n
i

i=0
such that
lim
i
n
i+1
n
i
n
i+2
n
i+1
= 0
and dene
1(^) =
_
A N : lim
i
[A [n
i
, n
i+1
)[
n
i+1
n
i
= 0
_
.
Theorem 6.1. A(1(^)) = 2

0
.
Proof. To begin the following claim will be established:
Claim 5. If g S(1(^)) then there is B 1(^) such that if j [n
i
, n
i+1
) B then
g(j) [n
i
, n
i+1
).
Proof. Let
B
+
=

_
i=0
n N : n
i
n < n
i+1
: g(n) n
i+1

and let
B

_
i=0
n N : n
i
n < n
i+1
: g(n) < n
i
.
MAXIMAL ABELIAN SUBGROUPS 35
If B
+
B

1(^) then the claim is proved. To begin suppose B

+
/ 1(^). Choose
> 0 and an innite Y N such that
[B
+
[n
i
, n
i+1
)[
n
i+1
n
i

for each i Y . By thinning out Y it may also be assumed that if i and j belong to Y
and i < j and m B
+
[n
i
, n
i+1
) then g(m) < n
j
. It follows that
g(B
+
) [n
i+1
, n
j
) = g
_
B
+
[n
i
, n
i+1
)
_
[n
i+1
, n
j
).
Therefore, if i < k < j then
[g(B
+
) [n
k
, n
k+1
)[
n
k+1
n
k

n
i+1
n
i
n
k+1
n
k
and so g(B
+
) 1(^) contradicting that g S(1(^)). A similar argument applied to
g
1
deals with B

.
Now suppose that G S(1(^)) is a maximal subset whose elements almost com-
mute modulo 1(^) and let g G F(1(^)). Using the Claim, there is B 1(^)
such that if j [n
i
, n
i+1
) B then j ,= g(j) [n
i
, n
i+1
). Let g
t
be a permutation such
that g
t
N B = g N B and g
t
[n
i
, n
i+1
) is a permutation of [n
i
, n
i+1
) for each
i. (This is possible since g [n
i
, n
i+1
) B [n
i
, n
i+1
) is one-to-one.) Note that g
t
belongs to the same coset of F(1(^)) as g. For any Z N let g
Z
be dened by
g
Z
(j) =
_
g
t
(j) if j [n
i
, n
i+1
) and i Z
j if j [n
i
, n
i+1
) B and i / Z
and note that g
Z
S(1(^)) for each Z. Moreover, there is some > 0 and an innite
X N such that
[m [n
i
, n
i+1
) : g
t
(m) ,= m[
n
i+1
n
i
is greater than for each i X. It follows that if Z and W are subsets of X and
[ZW[ =
0
then g
Z
and g
W
do not belong to the same coset of F(1(^)). Hence it
will suce to show that each g
Z
commutes with each member of G.
To this end, let Z N and suppose that h G and use the claim to nd C 1(^)
such that h(j) [n
i
, n
i+1
) for each i and each j [n
i
, n
i+1
) C. Since h and g
almost commute modulo 1(^) let D 1(^) be such that h(g(j)) = g(h(j)) for
j N D. Then let E = B h
1
(B) C D and note that E 1(^) since
h S(1(^)). It will be shown that g
Z
(h(j)) = h(g
Z
(j)) for each j N E. To see
this let j [n
i
, n
i+1
)E and suppose rst that i Z. In this case g
Z
(j) = g
t
(j) = g(j)
because j / B. Furthermore, since j / C, h(j) [n
i
, n
i+1
) and h(j) / B and hence
g(h(j)) = g
Z
(h(j)). Since j / D it follows that h(g(j)) = g(h(j)) and, hence, in this
case h(g
Z
(j)) = g
Z
(h(j)). If i / Z then g
Z
(j) = j because j / B and, since j / C,
h(j) [n
i
, n
i+1
). Therefore, since h(j) / B, g
Z
(h(j)) = h(j) = h(g
Z
(j)).
36 S. SHELAH AND J. STEPR

ANS
Theorem 6.2. a(1(^)) a.
Proof. Let / be a maximal almost disjoint family of size a. For A / dene A

iA
[n
i
, n
i+1
) and let /

= A

: A /. Then /

is maximal in T(N)/1(^).
It follows that a(1(^)) < A(1(^)) in any model of set theory where a ,= 2

0
.
7. Questions
It is well known that maximal almost disjoint families of subsets of N can not have
any nice denition; indeed, Mathias has shown that they can not be analytic [7].
The results of 2 which establish some similarity between A([N]
<
0
) and a raise the
following question.
Question 7.1. Is there an analytic, maximal, almost commuting subgroup of S ?
The lower bound of 3 is probably not optimal.
Question 7.2. Can the lower bound A(S/F) p of Theorem 3.2 be improved? Can
the tower invariant t serve as a lower bound?
For any function h : N R one can dene the summable ideal 1
h
to be the set
of all X N such that

xX
h(x) < . Observe that it is possible to modify the
proof of Theorem 6.1 in order to replace the ideal 1(^) by a summable ideal. In
particular, let n
i

i=0
be an increasing sequence of integers dened by n
i+1
n
i
= n
3
i
and let h be dened by h(j) = n
3
i
if n
i
j < n
i+1
. If g S(1
h
) and B
+
and B

are
dened as in the proof of Theorem 6.1 then it is easy to see that

jB
+
n
i
h(g(j))
[B
+
n
i
[n
3
i
n
2
i
and hence Claim 5 still holds as does the remainder of the argument
of Theorem 6.1. Hence A(1
h
) = 2

0
. This motivates the following question.
Question 7.3. For which functions h is it possible to improve Theorem 5.1 to show
that A(1
h
) =
1
< 2

0
in the model obtained by adding
1
Cohen reals?
Question 7.4. Are there functions h and g such that it is consistent that A(1
h
) <
A(1
g
) < 2

0
?
Question 7.5. Is it possible to characterize the summable ideals 1
h
such that A(1
h
) =
2

0
? Can the same be done for the F

ideals? What can be said of the Borel or analytic

ideals?
At some points in the proof of 5 the permutations constructed can be taken to be
almost commuting rather than just almost commuting modulo 1
1/x
. The following
question asks whether the argument can be strengthened throughout.
MAXIMAL ABELIAN SUBGROUPS 37
Question 7.6. Can Theorem 5.1 be improved to show that it is consistent with set
theory that 2

0
>
1
yet there is an almost commuting subgroup of S of cardinality

1
which is maximal with respect to commuting modulo 1
1/x
? Does this hold in the
Cohen model of Theorem 5.1?
The methods of 4 and 5 require that the subgroups constructed contain many
involutions. While the methods can be modied to produce groups with no elements
of order k for a xed k, the following questions seem more subtle.
Question 7.7. Can Theorem 4.2 be modied to assert that it is consistent that there
is a maximal, abelian, torsion free subgroup of S/F of size
1
and
1
< a ?
Question 7.8. Can Theorem 5.1 be modied to assert that it is consistent that there
is a maximal, abelian, torsion free subgroup of S(1
1/x
)/F(1
1/x
) of size
1
and
1
< 2

0
?
References
[1] J. L. Alperin, Jacinta Covington, and Dugald Macpherson. Automorphisms of quotients of sym-
metric groups. In Ordered groups and innite permutation groups, pages 231247. Kluwer Acad.
Publ., Dordrecht, 1996.
[2] Tomek Bartoszy nski and Haim Judah. Set Theory On the structure of the real line. A K
Peters, 1995.
[3] Murray G. Bell. On the combinatorial principle P(c). Fund. Math., 114(2):149157, 1981.
[4] J. M. Burns and B. Goldsmith. Maximal order abelian subgroups of symmetric groups. Bull.
London Math. Soc., 21(1):7072, 1989.
[5] Alan Dow. Two classes of Frechet-Urysohn spaces. Proc. Amer. Math. Soc., 108(1):241247,
1990.
[6] Ilijas Farah. Analytic quotients: theory of liftings for quotients over analytic ideals on the inte-
gers. Mem. Amer. Math. Soc., 148(702):xvi+177, 2000.
[7] A. R. D. Mathias. Happy families. Ann. Math. Logic, 12(1):59111, 1977.
[8] Jan Mycielski. Some theorems on connected compact groups. Bull. Acad. Polon. Sci. Cl. III,
5:1023, LXXXV, 1957.
[9] Jan Mycielski. Some properties of connected compact groups. Colloq. Math., 5:162166, 1958.
[10] Walter Rudin. Homogeneity problems in the theory of

Cech compactications. Duke Math. J.,
23:409419, 1956.
[11] J. Schreier and S. Ulam. Uber die permutationsgruppe der naturlichen zahlenfolge. Studia
Math., 14:134141, 1933.
[12] S. Shelah and J. K. Truss. On distinguishing quotients of symmetric groups. Ann. Pure Appl.
Logic, 97(1-3):4783, 1999.
[13] Saharon Shelah. Non-elementary proper forcing notions. Shelah [Sh:630].
[14] Saharon Shelah. First order theory of permutation groups. Israel J. Math., 14:149162, 1973.
[15] Juris Steprans. The almost disjointness cardinal invariant in the quotient algebra of the rationals
modulo the nowhere dense subsets. Real Anal. Exchange, 27(2):795800, 2001/02.
[16] Reinhard Winkler. On maximal abelian groups of maps. J. Austral. Math. Soc. Ser. A, 55(3):414
420, 1993.
38 S. SHELAH AND J. STEPR

ANS
Department of Mathematics, Rutgers University, Hill Center, Piscataway, New
Jersey, U.S.A. 08854-8019
Current address: Institute of Mathematics, Hebrew University, Givat Ram, Jerusalem 91904, Israel
E-mail address: shelah@math.rutgers.edu
Department of Mathematics, York University, 4700 Keele Street, North York,
Ontario, Canada M3J 1P3
Current address: Fields Institute, 222 College Street, Toronto, Canada M5T 3J1
E-mail address: steprans@yorku.ca