5

2
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
1
2
-
2
6







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
6


THE COFINALITY SPECTRUM OF
THE INFINITE SYMMETRIC GROUP
Saharon Shelah
and
Simon Thomas
Research partially supported by the BSF. Publication 524 of the rst author.
Research partially supported by NSF Grants.
Typeset by A
M
S-T
E
X
1
5
2
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
1
2
-
2
6







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
6


2 SAHARON SHELAH AND SIMON THOMAS
1. Introduction.
Suppose that G is a group that is not nitely generated. Then G can be written
as the union of a chain of proper subgroups. The conality spectrum of G, written
CF(S), is the set of regular cardinals such that G can be expressed as the union
of a chain of proper subgroups. The conality of G, written c(G), is the least
element of CF(G).
Throughout this paper, S will denote the group Sym() of all permutations
of the set of natural numbers. In [MN], Macpherson and Neumann proved that
c(S) >
o
. In [ST1] and [ST2], the possibilities for the value of c(S) were studied.
In particular, it was shown that it is consistent that c(S) and 2

o
can be any two
prescribed regular uncountable cardinals, subject only to the obvious requirement
that c(S) 2

o
. In this paper, we shall begin the study of the possibilities for the
set CF(S).
There is one obvious constraint on the set CF(S), arising from the fact that
S can be expressed as the union of a chain of 2

o
proper subgroups; namely, that
cf(2

o
) CF(S). Initially it is dicult to think of any other constraints on CF(S).
And we shall show that it is consistent that CF(S) is quite a bizarre set of cardinals.
For example, the following result is a special case of our main theorem.
Theorem 1.1. Let T be any subset of 0. Then it is consistent that

n
CF(S) if and only if n T.
After seeing this result, the reader might suspect that it is consistent that CF(S)
is an arbitrarily prescribed set of regular uncountable cardinals, subject only to the
above mentioned constraint. However, this is not the case.
Theorem 1.2. If
n
CF(S) for all n 0, then
+1
CF(S).
(Of course, this result is only interesting when 2

0
>
+1
.) In Section 2, we
shall use pcf theory to prove Theorem 1.2, together with some further results which
restrict the possibilities for CF(S). In Section 3, we shall prove the following result.
Theorem 1.3. Suppose that V GCH. Let C be a set of regular uncountable
cardinals which satises the following conditions.
(1.4)
(a) C contains a maximum element.
(b) If is an inaccessible cardinal such that = sup(C ), then C.
(c) If is a singular cardinal such that = sup(C ), then
+
C.
Then there exists a c.c.c notion of forcing P such that V
P
CF(S) = C.
This is not the best possible result. In particular, clause (1.4)(c) can be improved
so that we gain a little more control over what occurs at successors of singular
cardinals. This matter will be discussed more fully at the end of Section 2. Also
clause (1.4)(a) is not a necessary condition. For example, let V GCH and let
C =
+1
<
1
. At the end of Section 3, we shall show that if is any singular
cardinal such that cf() C, then there exists a c.c.c notion of forcing P such that
V
P
CF(S) = C and 2

o
= . In particular, 2

o
cannot be bounded in terms of
the set CF(S).
5
2
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
1
2
-
2
6







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
6


THE COFINALITY SPECTRUM OF THE INFINITE SYMMETRIC GROUP 3
In this paper, we have made no attempt to control what occurs at inaccessible
cardinals such that = sup(C ). We intend to deal with this matter in a
second paper, which is in preparation. In this second paper, we also hope to give
a complete characterisation of those sets C for which there exists a c.c.c notion of
forcing P such that V
P
CF(S) = C.
Our notation mainly follows that of Kunen [K]. Thus if P is a notion of forcing
and p, q P, then q p means that q is a strengthening of p. If V is the ground
model, then we often denote the generic extension by V
P
if we do not wish to specify
a particular generic lter G P. If we want to emphasize that the term t is to be
interpreted in the model M of ZFC, then we write t
M
; for example, Sym()
M
. If
A , then S
(A)
denotes the pointwise stabilizer of A. Fin() denotes the subgroup
of elements S such that the set n < (n) ,= n is nite. If , S, then we
dene =

if and only if
1
Fin().
2. Some applications of pcf theory.
Let
i
i I) be an indexed set of regular cardinals. Then
iI

i
denotes the
set of all functions f such that dom f = I and f(i)
i
for all i I. If T is
a lter on I and 1 is the dual ideal, then we write either
iI

i
/
F
or
iI

i
/
I
for
the corresponding reduced product. We shall usually prefer to work with functions
f
iI

i
rather than with the corresponding equivalence classes in
iI

i
/
I
. For
f, g,
iI

i
, we dene
f
I
g i i If(i) > g(i) 1
f <
I
g i i If(i) g(i) 1.
We shall sometimes write f
F
g, f <
F
g instead of f
I
g, f <
I
g respectively.
If 1 = , then we shall write f g, f < g. Suppose that there exists a regular
cardinal and a sequence f

[ < ) of elements of
iI

i
such that
(a) if < < , then f

<
I
f

; and
(b) for all h
iI

i
, there exists < such that h <
I
f

.
Then we say that is the true conality of
iI

i
/
I
, and write tcf

iI

i
/
I

= .
Furthermore, we say that f

[ < ) witnesses that tcf

iI

i
/
I

= . For
example, if T is an ultralter on I, then
iI

i
/
D
is a linearly ordered set and hence
has a true conality. A cardinal is a possible conality of
iI

i
if there exists an
ultralter T on I such that tcf

iI

i
/
D

= . The set of all possible conalities

of
iI

i
is pcf

iI

.
In recent years, Shelah has developed a deep and beautiful theory of the structure
of pcf

iI

when [I[ < min

i
[i I. A thorough development of pcf theory
and an account of many of its applications can be found in [Sh-g]. [BM] is a self-
contained survey of the basic elements of pcf theory. In this section of the paper, we
5
2
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
1
2
-
2
6







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
6


4 SAHARON SHELAH AND SIMON THOMAS
shall see that pcf theory imposes a number of constraints on the possible structure
of CF(S). (Whenever it is possible, we shall give references to both [Sh-g] and
[BM] for the results in pcf theory that we use.)
Theorem 2.1. Suppose that
n
[n < ) is a strictly increasing sequence of cardi-
nals such that
n
CF(S) for all n < . Let T be a nonprincipal ultralter on ,
and let tcf

n<

n
/
D

= . Then CF(S).
Proof. For each n < , express S =

i<
n
G
n
i
as the union of a chain of
n
proper subgroups. Let f

[ < ) be a sequence in
n<

n
which witnesses that
tcf

n<

n
/
D

= . For each < , let H

be the set of all g S such that

n < [g G
n
f

(n)
T. Then it is easily checked that H

is a subgroup of S,
and that H

for all < < . Suppose that g S is an arbitrary element.

Dene f
n<

n
by f(n) = mini[g G
n
i
. Then there exists < such that
f <
D
f

. Hence g H

. Thus S =

<
H

.
So it suces to prove that H

is a proper subgroup of S for each < .

Fix some < . Lemma 2.4 [MN] implies that for each n < , i <
n
and
X []

, the setwise stabilizer of X in G

n
i
does not induce Sym (X) on X.
Express =

n<
X
n
as the disjoint union of countably many innite subsets X
n
.
For each n < , choose
n
Sym (X
n
) such that g X
n
,=
n
for all g G
n
f

(n)
.
Then =

n<

n
S H

Proof of Theorem 1.2.

By [Sh-g, II 1.5] (or see [BM, 2.1]), there exists an ultralter T on such that
tcf

n<

n
/
D

=
+1
.

If we assume MA

, then we can obtain the analogous result for cardinals such

that
o
< < 2

o
. (In Section 3, we shall prove that the following result cannot
be proved in ZFC.)
Theorem 2.2 (MA

). Suppose that

[ < ) is a strictly increasing sequence

of cardinals such that

CF(S) for all < . Let T be a nonprincipal ultralter

on , and let tcf

<

/
D

= . Then CF(S).
Proof. For each < , express S =

i<

i
as the union of a chain of

proper subgroups. Let f

[ < ) be a sequence in
<

which witnesses that

tcf

<

/
D

= . For each < , let H

be the set of all g S such that

< [g G

()
T. Arguing as in the proof of Theorem 2.1, it is easily
checked that H

[ < ) is a chain of subgroups such that S =

<
H

.
5
2
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
1
2
-
2
6







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
6


THE COFINALITY SPECTRUM OF THE INFINITE SYMMETRIC GROUP 5
Thus it suces to prove that H

is a proper subgroup of S for each < . Fix

some < . Suppose that we can nd an element g S

<
G

()
.
Then clearly g / H

. But the existence of such an element g is an immediate

consequence of the following theorem.

Theorem 2.3 (MA

). Suppose that for each < , S =

i<

i
is the union of
the chain of proper subgroups H

i
. Then for each f
<

, S ,=

<
H

f()
.
Remark 2.4. In [ST 1], it was shown that MA

implies that c(S) > . This result

is an easy consequence of Theorem 2.3.
Remark 2.5. In [MN], Macpherson and Neumann proved that if H
n
[n <
is an arbitrary set of proper subgroups of S, then S ,=

n<
H
n
. It is an open
question whether MA

implies the analogous statement for cardinals such that

o
< < 2

o
. Regard S as a Polish space in the usual way. Then the proof of
Theorem 2.3 shows that the following result holds.
Theorem 2.6 (MA

). Suppose that for each < , H

is a nonmeagre proper
subgroup of S. Then S ,=

<
H

Unfortunately there exist maximal subgroups H of S such that H is meagre. For

example, let =
1

2
be a partition of into two innite pieces. Let
H = g S

[g[
1
]
i
[ <
o
for some i 1, 2.
(Here denotes the symmetric dierence.) Then H is a maximal subgroup of S;
and it is easily checked that H is meagre.
Proof of Theorem 2.3 (MA

). We shall make use of the technique of generic se-

quences of elements of S, as developed in [HHLSh]. (The slight dierences in
notation between this paper and [HHLSh] arise from the fact that permutations
act on the left in this paper.)
Denition 2.7. A nite sequence g
1
, , g
n
) S
n
is generic if the following two
conditions hold.
(1) For all A []
<
, there exists A B []
<
such that g
i
[B] = B for all
1 i n.
(2) Suppose that A []
<
and that g
i
[A] = A for all 1 i n. Suppose
further that A B []
<
and that h
i
Sym(B) extends g
i
A for all
1 i n. Then there exists S
(A)
such that g
i

1
extends h
i
for all
1 i n.
Claim 2.8. If g
1
, ..., g
n
), h
1
, ..., h
n
) S
n
are generic, then there exists f S
such that fg
i
f
1
= h
i
for all 1 i n.
Proof of Claim 2.8. This follows from Proposition 2.3 [HHLSh].

From now on, regard S as a Polish space in the usual way.

5
2
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
1
2
-
2
6







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
6


6 SAHARON SHELAH AND SIMON THOMAS
Claim 2.9. The set g
1
, ..., g
n
) S
n

g
1
, ..., g
n
) is generic is comeagre in S
n
in
the product topology.
Proof of Claim 2.9. This follows from Theorem 2.9 [HHLSh].

Claim 2.10. If g
1
, ..., g
n+1
) S
n+1
is generic, then for each A []
<
, m A
and 1 n + 1, the following condition holds.
(2.11)
A,m,
Let = i

1 i n + 1, i ,= . If g
i
[A] = A for all i , then there
exists B [ A]
<
such that
(a) m B;
(b) g
i
[B] = B for all i ;
(c) g

[A B] = A B;
(d) for all Sym (), there exists Sym (B) such that (g
i
B)
1
=
g
(i)
B for all i .
Proof of Claim 2.10. For each A []
<
, m A and 1 n + 1,
let C
n+1
(A, m, ) consist of the sequences g
1
, ..., g
n+1
) S
n+1
which satisfy
(2.11)
A,m,
. Then it is easily checked that C
n+1
(A, m, ) is open and dense in
S
n+1
. Hence C
n+1
=

A,m,
C
n+1
(A, m, ) is comeagre in S
n+1
. Claim 2.9 implies
that there exists a generic sequence g
1
, ..., g
n+1
) C
n+1
. So the result follows
easily from Claim 2.8.

Denition 2.12. If is an innite ordinal, then the sequence g

i

i < ) of elements
of S is generic if for every nite subsequence i
1
< ... < i
n
< , g
i
1
, ..., g
i
n
) is generic.
We have now developed enough of the theory of generic sequences to allow us
to begin the proof of Theorem 2.3. Consider the chains of proper subgroups,
S =

i<

i
for < . We can assume that Fin() H

o
for all < . Let
f
<

. We must nd an element S

<
H

f()
. We shall begin by
inductively constructing a generic sequence of elements of S.
g
o
o
, g
1
o
, ..., g
o

, g
1

, ...)
<
such that for all < , there exist f()

<

such that g
o

and
g
1

/ H

. Then we shall nd an element S such that g

o

1
=

g
1

for all
< . This implies that /

<
H

<
H

f()
.
Suppose that we have constructed g
o

, g
1

for < . For each nite subsequence

g of g
o

, g
1

< ), the set h S

gh is generic is comeagre in S. (See[HHLSh],

page 216.) Since MA

implies that the union of meagre subsets of a Polish space

is meagre, the set
h S

g
o

, g
1

[ < )h is generic
is also comeagre in S. So we can choose a suitable g
o

and f()

<

with
g
o

. The set
C = h S

g
o

, g
1

< )g
o

h is generic
5
2
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
1
2
-
2
6







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
6


THE COFINALITY SPECTRUM OF THE INFINITE SYMMETRIC GROUP 7
is also comeagre in S. Since H

is a proper subgroup of S, we have that

C H

,= . (If not, then H

is comeagre and hence so are each of its cosets

in S. As any two comeagre subsets of S intersect, this is impossible.) Hence we
can choose a suitable g
1

C H

. Thus the desired generic sequence can be

constructed.
Lemma 2.13. Let g
o

, g
1

< ) be a generic sequence of elements of S. Then

there exists a -centred notion of forcing P such that

P
There exists Sym () such that g
o

1
=

g
1

for all < .

Proof of Lemma 2.13. Let P consist of the conditions p = h, F) such that
(a) there exists A []
<
such that h Sym(A);
(b) F []
<
;
(c) for each F and 0, 1, g

[A] = A.
We dene h
2
, F
2
) h
1
, F
1
) i the following two conditions hold.
(1) h
1
h
2
and F
1
F
2
.
(2) Let B = dom h
2
dom h
1
and let = h
2
B. Then (g
o

B)
1
= g
1

B
for each F
1
.
Clearly P is centered. Claim 2.10 implies that each of the sets
D
m
= h, F)

m dom h , m <
and
E

= h, F)

F , < ,
are dense in P. The result follows.
This completes the proof of Theorem 2.3.
The following theorem goes some way towards explaining why we have assumed
that C satises condition (1.4)(c) in the statement of Theorem 1.3. (We will discuss
this matter fully after we have proved Theorem 2.15.)
Denition 2.14. If is a limit ordinal, then J
bd

is the ideal on dened by

J
bd

= B

There exists i < such that B i.

Theorem 2.15. Let be a regular cardinal, and suppose that

[ < ) is a
strictly increasing sequence of cardinals such that

CF(S) for all < .

Suppose further that tcf

<

/
J
bd

= . Then either CF(S) or CF(S).

Proof. Suppose that / CF(S). For each < , express S =

i<

i
as the union
of a chain of

proper subgroups. Let f

[ < ) be a sequence in
<

which
witnesses that tcf

<

/
J
bd

= . For each < , let G

be the set of all g S

5
2
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
1
2
-
2
6







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
6


8 SAHARON SHELAH AND SIMON THOMAS
such that < [g G

()
J
bd

. Arguing as before, it is easily checked

that G

[ < ) is a chain of subgroups such that S =

<
G

.
Thus it suces to prove that G

is a proper subgroup of S for each < . So

suppose that G

= S for some < . For each < , dene

H

=

G

()
[ < . Then H

[ < ) is a chain of subgroups such that

S =

<
H

. If < , then H

()
and so H

is a proper subgroup of S.
But this contradicts the assumption that / CF(S).

Suppose that V GCH, and that is a singular cardinal. Let

i
[i < ) be
the strictly increasing enumeration of all regular uncountable cardinals such that
< . Let T =
i<

i
. Then [T[ =
+
. Now let P be any c.c.c. notion of forcing.
From now on, we shall work in V
P
. Since P is c.c.c., for each g
i<

i
, there exists
f T such that g f. Suppose now that

[ < ) is an increasing subsequence

of
i
[i < ) such that [[ <
o
and sup
<

= . Let
T

= f
<

There exists h T such that f h.

Then for all g
<

, there exists f T

such that g f. This im-

plies that max(pcf(
<

)) =
+
. By [Sh-g,I] (or see [BM,4.3]), we obtain that
tcf(
<

/
J
bd

) =
+
. In summary, we have shown that the following statement is
true in V
P
.
The Strong Hypothesis (2.16). Let be a limit ordinal, and let

[ < )
be a strictly increasing sequence of regular cardinals such that [ <
o
. Then
tcf

<

/
J
bd

= (sup
<

)
+
.
In particular, using Theorem 2.15 and the Strong Hypothesis, we see that the
following statement is true in V
P
.
() If is a singular cardinal such that = sup(CF(S) ), then either
cf() CF(S) or
+
CF(S).
This suggests that we might try to replace condition (1.4)(c) of Theorem 1.3 by
the following condition.
(1.4)(c)

If is a singular cardinal such that = sup(C ), then either

cf() C or
+
C.
However, Theorem 2.19 shows that this cannot be done. For example, Theorem
2.19 implies that if
C =
1

+1

<
2
, cf() =

2
+1
,
then there does not exist a c.c.c. notion of forcing P such that V
P
CF(S) = C.
5
2
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
1
2
-
2
6







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
6


THE COFINALITY SPECTRUM OF THE INFINITE SYMMETRIC GROUP 9
Remark 2.17. The Strong Hypothesis is usually taken to be the following apparently
weaker statement.
(2.18) For all singular cardinals , pp() =
+
.
(For the denition of pp(), see [Sh-400a].) However, Shelah [Sh-420, 6.3 (1)] has
shown that (2.16) and (2.18) are equivalent.
Theorem 2.19 (The Strong Hypothesis). Let be a regular uncountable car-
dinal, and suppose that

[ < ) is a strictly increasing sequence of cardinals

such that

CF(S) for all < . Suppose further that

(a) <
o
;
(b) E = <

lim, (sup
<

)
+
/ CF(S) is a stationary subset of .
Then CF(S).
Proof. For each < , express S =

i<

i
as the union of a chain of

proper subgroups. For each E, let

= sup
<

. By the Strong Hypothesis,

tcf

<

/
J
bd

=
+

. Let f

[ <
+

) be a sequence in
<

which witnesses
that tcf

<

/
J
bd

=
+

. For each <

+

, let H

be the set of all g S

such that < [g G

()
J
bd

. Once again, it is easily checked that

H

[ <
+

) is a chain of subgroups such that S =

<
+

. Since
+

/ CF(S),
there exists () <
+

such that H

()
= S.
Since <
o
, there exists f
<

such that f() > supf

()
()[ < E
for all < . Let g S. Then for each E, g H

()
; and so there exists
(g, ) < such that g G

()
()
G

f()
for all (g, ) < . By Fodors
Theorem, there exists an ordinal (g) < and a stationary subset D of E such
that (g, ) = (g) for all D. This means that g

G

f()
(g) < .
For each < , let

=

G

f()
< . Then

[ < ) is a chain
of subgroups such that S =

<

. Finally note that

f()
, and so

is a
proper subgroup of S for all < . Thus CF(S).

3. The main theorem. In this section, we shall prove Thoerem 1.3. Our notation
generally follows that of Kunen [K]. We shall only be using nite support iterations.
An iteration of length will be written as P

,

Q

[ , < ), where P

is the
result of the rst stages of the iteration, and for each < there is some P

-name

such that

is a partial ordering
and P
+1
is isomorphic to P

. If p P

, then supt(p) denotes the support of

p.
5
2
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
1
2
-
2
6







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
6


10 SAHARON SHELAH AND SIMON THOMAS
There is one important dierence between our notation and that of Kunen.
Unlike Kunen, we shall not use V
P
to denote the class of P-names for a notion of
forcing P. Instead we are using V
P
to denote the generic extension, when we do not
wish to specify a particular generic lter G P. Normally it would be harmless to
use V
P
in both of the above senses, but there is a point in this section where this
notational ambiguity could be genuinely confusing. Suppose that Q is an arbitrary
suborder of P. Then the class of Q-names is always a subclass of the class of P-
names. (Of course, a Q-name might have very dierent properties when regarded
as a P-name. For example, it is possible that
Q
is a function, whilst
P
is a
function.) However, we will not always have that V
Q
V
P
; where this means that
V [G Q] V [G] for some unspecied generic lter G P.
Denition 3.1. Let Q be a suborder of P. Q is a complete suborder of P, written
Q P, if the following two conditions hold.
1. If q
1
, q
2
Q and there exists p P such that p q
1
, q
2
, then there exists
r Q such that r q
1
, q
2
.
2. For all p P, there exists q Q such that whenever q

Q satises q

q,
then q

and p are compatible in P. (We say that q is a reduction of p to Q.)

It is wellknown that if Q P, then V
Q
V
P
; and we shall only write V
Q
V
P
when Q P.
We are now ready to explain the idea behind the proof of Theorem 1.3. Let
V GCH, and let C be a set of regular uncountable cardinals which contains a
maximum element . We seek a c.c.c. P such that V
P
2

= CF(S) = C. The
easiest part of our task is to ensure that V
P
C CF(S). We shall accomplish
this by constructing P so that the following property holds for each C.
(3.2)

There exists a sequence P

[ < ) V of suborders of P such that

(a) if < < , then P

P;
(b) for each Sym()
V
P
, there exists < such that Sym()
V
P

;
(c) for each < , there exists Sym()
V
P
Sym()
V
P

.
The harder part is to ensure that V
P
CF(S) C. This includes the require-
ment that (3.2)

fails for every / C. So, roughly speaking, we are seeking a

c.c.c. P which can be regarded as a kind of iteration of length precisely when
C. We shall use the technique of Section 3 [Sh-288] to construct such a notion
of forcing P.
Denition 3.3. Let a
i
i < ) be a sequence of subsets of . We say that b is
closed for a
i
[i < ) if a
i
b for all i b.
Denition 3.4. Let ( be the class of all sequences

Q = P
i
,

Q
j
, a
j
i , j < )
for some which satisfy the following conditions. (We say that

Q has length and
write = lg (

Q).)
(a) a
i
i.
5
2
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
1
2
-
2
6







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
6


THE COFINALITY SPECTRUM OF THE INFINITE SYMMETRIC GROUP 11
(b) a
i
is closed for a
j
[j < i).
(c) P
i
is a notion of forcing and

Q
j
is a P
j
-name such that
P
j

Q
j
is a c.c.c.
partial order.
(d) P
i
,

Q
j
[i , j < ) is a nite support iteration.
(e) For each j < , dene the suborder P

a
j
of P
j
inductively by
P

a
j
= p P
j
[ supt(p) a
j
and p(k) is a P

a
k
name for all k supt(p).
Then

Q
j
is a P

a
j
-name. (At this stage, we do not know whether P

a
j
is a complete
suborder of P
j
. It is for this reason that we are being careful with our notation.
However, we shall soon see that P

a
j
P
j
, and then we can relax again.)
Denition 3.5. Let

Q ( be as above, so that = lg (

Q).
(a) We say that b is closed for

Q if b is closed for a
j
[j < ).
(b) If b is closed for

Q, then we dene P

b
= p P

supt(p) b and p(k)

is a P

a
k
-name for all k supt(p).
If < , then we identify P

with the corresponding complete suborder of P

in the usual way. If b , then p b denotes the -sequence dened by

(p b)() = p() if b
= 1

otherwise
Lemma 3.6. Let

Q ( and let =lg (

Q). Suppose that b c , and that
b and c are closed for

Q.
(1) is closed for

Q, and P

= P

.
(2) If p P

and i supt(p), then p a

i
p(i)

Q
i
.
(3) Suppose that p, q P

and p q. If i supt(q), then p a

i
p(i) q(i).
(4) If p P

c
, then p b P

b
.
(5) Suppose that p P

c
, q P

b
and p q. Then p b q.
(6) Suppose that p P

c
, q P

and p q c.
Dene the -sequence r by
r() = p() if c
= q() otherwise.
Then r P

and r p, q.
(7) P

c
P

.
Proof. This is left as a straightforward exercise for the reader.

5
2
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
1
2
-
2
6







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
6


12 SAHARON SHELAH AND SIMON THOMAS
Lemma 3.7. Let

Q ( and let = lg (

Q). Suppose that b is closed for

Q
and that i b.
(1) c = b i and c i are closed for

Q.
(2) P

b
P

c
P

c{i}
P

.
(3) P

c{i}
is isomorphic to P

c


Q
i
.
Proof. Once again left to the reader.

Now we are ready to begin the proof of Theorem 1.3. Suppose that V GCH,
and let C be a set of regular uncountable cardinals which satises the following
conditions.
(1.4)
(a) C contains a maximum element, say .
(b) If is an inaccessible cardinal such that = sup(C ), then C.
(c) If is a singular cardinal such that = sup(C ), then
+
C.
Denition 3.8.
(a) C denotes the set of all functions f such that dom f = C and f()
for all C.
(b) T
C
is the set of all functions f C which satisfy the following condition.
() If is an inaccessible cardinal such that = sup(C ), then there exists
< such that f() = 0 for all C .
Denition 3.9. In V , we dene a sequence
P
i
,

Q
j
, f
j
i , j < )
such that the following conditions are satised.
(a) f
i
T
C
.
(b) Let a
i
= j < if
j
f
i
. Then

Q = P
i
,

Q
j
, a
j
[i , j < ) (.
(c) For each f T
C
, there exists a conal set of ordinals j < such that
f
j
= f.
(d) Suppose that i < and that

Q is a P

a
i
-name with [

Q[ < . Then there

exists i < j < such that
(1) f
j
= f
i
, and so a
i
a
j
;
(2) if
P
j

Q is c.c.c., then

Q
j
=

Q.
We shall prove that V
P

CF(S) = C. From now on, we shall work inside V

P

.
Denition 3.10. If b is closed for

Q, then S
b
= Sym()
V
P

b
.
First we shall show that C CF(S). Fix some C. For each < , let
b

= i < [f
i
() . Clearly b

is closed for

Q; and if < < , then b

.
Thus S
b

< ) is a chain of subgroups of S.

5
2
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
1
2
-
2
6







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
6


THE COFINALITY SPECTRUM OF THE INFINITE SYMMETRIC GROUP 13
Lemma 3.11. For each < , S
b

is a proper subgroup of S.
Proof. Let < and let i < satisfy f
i
() > . Let Q be the partial order of
nite injective functions q : , and let

Q be the canonical P

a
i
-name for Q.
Then there exists i < j < such that f
j
= f
i
and

Q
j
=

Q. Clearly j / b

. Let
c = b

j. By Lemma 3.7,

Q
j
adjoins a permutation of such that / V
P

c
. It
follows that / S
b

.
Lemma 3.12.
S =

<
S
b

.
Proof. Let S. Let g be a nice P

-name for . (Remember that P

= P

.) Thus
there exist antichains A
,m
of P

for each , m) such that

g =

,m
, m) A
,m
. Let

supt(p)[p

,m
A
,m
=
k
[k < . Let
= supf

k
()[k < . Then p P

for each p

,m
A
,m
, and so g is a nice
P

-name. Hence S
b

This completes the proof of the following result.

Lemma 3.13. If C, then CF(S).

To complete the proof of Theorem 1.3, we must show that if / C, then

/ CF(S). We shall make use of the following easy observation.
Lemma 3.14. Let M ZFC, and let g

[ < ) M be a generic sequence

of elements of Sym(). Let Q be the partial order of nite injective functions
q : , and let M
Q
be the Q-generic permutation. Then for all
h Sym()
M
, g

[ < )h is generic.
Proof. For each nite subsequence
1
< <
n
< , the set
C(
1
, ,
n
) = Sym()[g

1
, , g

n
) is generic is comeagre in Sym().
Hence h
1
C(
1
, ,
n
) is also comeagre for each h Sym(). So for each
h Sym()
M
, h
1
C(
1
, ,
n
). The result follows.

Lemma 3.15. Suppose that < and that g

[ < ) is a generic sequence of

elements of Sym(). If H is any proper subgroups of Sym(), then there exists a
permutation / H such that g

[ < ) is generic.
Proof. Let h Sym()H. Then there exists i < such that h, g

[ < ) V
P
i
.
There exists i < j < such that

Q
j
is the canonical P

a
j
-name for the partial order
Q of nite injective functions q : . By Lemma 3.14, there exists a permutation
V
P
j+1
such that both g

[ < ) and g

[ < )h are generic. Clearly

either / H or h / H.

Now x some / C, and suppose that CF(S). It is easily checked that

2

o
= , and so we can suppose that is a regular uncountable cardinal such that
5
2
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
1
2
-
2
6







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
6


14 SAHARON SHELAH AND SIMON THOMAS
< . Express S =

<
G

as the union of a chain of proper subgroups. We

can suppose that Fin() G
o
. Using Lemma 3.15, we can inductively construct a
generic sequence of elements of S
g
o
o
, g
1
o
, , g
o

, g
1

, )
<
such that for each < , there exists

< such that g

o

and g
1

/ G

.
Lemma 3.16. There exists a subset X []

and an ordinal < such that

g
o

, g
1

X) V
P

.
Proof. For each < and 0, 1, let g

be a nice P

-name for g

. Thus there
exist antichains A
,
,m
of P

for each , m) such that

g

,m
, m) A
,
,m
.
For each < , let

supt(p)[p

,m
A
,o
,m


,m
A
,1
,m
=

k
[k < . Dene
h

T
C
by h

() = supf

k
()[k < for each C.
It is easily checked that there are less than possibilities for the restriction
h

C . (This calculation is the only point in the proof of Theorem 1.3 where
we make use of the hypothesis that C satises conditions (1.4)(b) and (1.4)(c).)
Hence there exists X []

such that h

C = h

C for all , X.
Dene the function f C by f C = h

C , where X, and
f() = suph

()[ X for each C . Then it is easily checked that

f T
C
; and clearly f

k
h

f for all X and k < . Now choose

> sup

k
[ X, k < such that f

= f. If X and 0, 1, then
p P

for each p

,m
A
,
,m
; and hence g

is a nice P

-name. It follows that

g
o

, g
1

X) V
P

By Lemma 2.13, there exists a -centred Q V

P

such that

Q
There exists Sym() such that g
o

1
=

g
1

for all X.
Let

Q be a P

-name for Q. Then there exists < < such that f

= f

and

=

Q. Hence there exists S such that g
o

1
=

g
1

for all X. But

this implies that /

<
G

, which is a contradiction. This completes the proof of

Theorem 1.3.
By modifying the choice of the set T
C
of functions, we can obtain some inter-
esting variants of Theorem 1.3. For example, the following theorem shows that
Theorem 2.2 cannot be proved in ZFC. (Of course, it also shows that (1.4)(c) is not
a necessary condition in Theorem 1.3.)
5
2
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
1
2
-
2
6







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
6


THE COFINALITY SPECTRUM OF THE INFINITE SYMMETRIC GROUP 15
Theorem 3.17. Suppose that V GCH and that >

1
+1
is regular. Let
C =
+1
<
1
. Then there exists a c.c.c. notion of forcing P such that
V
P
CF(S) = C.
Proof. The proof is almost identical to that of Theorem 1.3. The only change is
that we use the set of functions
T

C
= f C[ There exists <
1
such that f(
+1
) = 0 for all <
1

in the denition of P

. This ensures that the counting argument in the analogue of

Lemma 3.16 goes through.

Using some more pcf theory, we can prove the following result.
Theorem 3.18. Suppose that V satises the following statements.
(a) 2

n
=
n+1
for all n < .
(b) 2

=
+1
for some < <
1
.
(c) 2

=
+1
for all .
Let T []

and let be a regular cardinal such that

+1
. Let C =
pcf(
nT

n
) . Then there exists a c.c.c. notion of forcing P such that V
P

CF(S) = C.
Proof. Again we argue as in the proof of Theorem 1.3. This time we use the set
of functions, T
#
C
=
nT

n
, in the denition of P

. Examining the proof of Lemma

3.16, we see that it is enough to prove that the following statement holds for each
regular uncountable / C.
(3.19)

If h

[ < ) is a sequence in
nT

n
, then there exists X []

and an f
nT

n
such thath

f for all X.
This is easy if <

. If >

, then (3.19)

is a consequence of the following

result.
Theorem 3.20. Let
i
[i I be a set of regular cardinals such that
min
i
[i I > [I[. Let be a regular cardinal such that > 2
|I|
and
/ pcf(
iI

i
). If h

[ < ) is a sequence in
iI

i
, then there exists X []

and f
iI

i
such that h

f for all X.
Proof. This is included in the proof of [Sh-g, II 3.1]. (More information on this topic
is given in [Sh-513, Section 5]. Also [Sh-430, 6.6D] gives even more information
under the hypothesis that 2
|I|
< min
i
[i I.) Alternatively, argue as in the
proof of [BM,7.11].

It is known that, assuming the consistency of a suitable large cardinal hypothesis,

for each < <
1
there exists a universe which satises the hypotheses of
Theorem 3.18. (See [GM].) Thus the following result shows that Theorem 1.2
cannot be substantially improved in ZFC.
5
2
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
1
2
-
2
6







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
6


16 SAHARON SHELAH AND SIMON THOMAS
Corollary 3.21. Suppose that V satises the hypotheses of Theorem 3.18 with
respect to some < <
1
. Then for each and
+1
, there exists
a set T []

and a c.c.c. notion of forcing P such that

V
P
CF(S) =
n
n T
+1
.
In particular, if < , then
V
P

+1
/ CF(S).
Proof. With the above hypotheses, [Sh-g,VIII] implies that there exists T []

such that tcf

nT

n
/
J
bd

=
+1
. It follows that
pcf(
nT

n
) =
n
[n T
+1
. So the result is a consequence of Theorem
3.18.

Finally we shall show that (1.4)(a) is not a necessary condition in Theorem 1.3,
and that 2

o
cannot be bounded in terms of the set CF(S).
Theorem 3.22. Suppose that V GCH and that C =
+1
<
1
. If is any
singular cardinal such that cf() C, then there exists a c.c.c notion of forcing P
such that V
P
CF(S) = C and 2

o
= .
Proof. Let be a singular cardinal such that cf() C. Let

[ < cf())
be a strictly increasing sequence of regular cardinals such that
0
=

1
+1
and
sup
<cf()

= . Let
T

C
= f C[ There exists <
1
such that f(
+1
) = 0 for all <
1
.
In V, we dene a sequence P
i
,

Q
j
, f
j
i , j < ) such that the following conditions
are satised.
(a) f
i
T

C
.
(b) Let a
i
= j < if
j
f
i
. Then

Q = P
i
,

Q
j
, a
j
[i , j < ) (.
(c) For each f T

C
and < cf(), there exists a conal set of ordinals j <

such that f
j
= f.
(d) Suppose that < cf(), i <

and that

Q is a P

a
i
-name with [

Q[ <

.
Then there exists i < j <

such that
(1) f
j
= f
i
, and so a
i
a
j
;
(2) if
P
j

Q is c.c.c., then

Q
j
=

Q.
Clearly V
P

0
= . Arguing as in the proof of Lemma 3.13, we see that
V
P

C CF(S). From now on, we shall work inside V

P

. Let be a regular
cardinal such that

1
+1
< . Suppose that we can express S =

<
G

as
the union of a chain of proper subgroups. For each < , choose an element
h

G G

. Then there exists a subset I []

and an ordinal < cf() such

5
2
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
1
2
-
2
6







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
6


THE COFINALITY SPECTRUM OF THE INFINITE SYMMETRIC GROUP 17
that h

I) V
P

and

. In V
P

, we can inductively construct a generic

sequence of elements of S
g
0
0
, g
1
0
, , g
0

, g
1

, )
<
such that for each <
(1) there exists

< such that g

0

and g
1

/ G

; and
(2) there exists

<
+1
such that g
0

, g
1

[ < ) V
P
i

.
For suppose that g
0

, g
1

[ < ) has been dened. By Lemma 3.14, there exists

i

< j <
+1
and g
0

V
P
j
such that g
0

, g
1

[ < )g
0

is generic. Choose

I
such that

< and g
0

. By a second application of Lemma 3.14, there

exists j < i
+1
<
+1
and V
P
i
+1
such that both g
0

, g
1

[ < )g
0

and
g
0

, g
1

[ < )g
0

are generic. Clearly either / G

or h

/ G

. Hence
we can also nd a suitable g
1

.
There exists a subset J []

and an ordinal < cf() such that

g
0

, g
1

[ J) V
P

and

. Arguing as in the proofs of Theorems 1.3 and

3.17, there exists V
P

+1
such that g
0

1
=

g
1

for all J. This is a

contradiction.

5
2
4


r
e
v
i
s
i
o
n
:
1
9
9
4
-
1
2
-
2
6







m
o
d
i
f
i
e
d
:
1
9
9
6
-
0
3
-
1
6


18 SAHARON SHELAH AND SIMON THOMAS
References
[BM] M. R. Burke and M. Magidor, Shelahs pcf theory and its applications, Ann. Pure Appl.
Logic 50 (1990), 207254.
[GM] M. Gitik and M. Magidor, The singular cardinal hypothesis revisited, in Set Theory of
the Continuum, 26 , Mathematical Sciences Research Institute Publications, (ed. H.
Judah, W. Just and H. Woodin), 1992, pp. 243279, Springer Verlag.
[HHLSh] W. Hodges, I. Hodkinson, D. Lascar and S. Shelah, The small index property for -
stable -categorical structures and for the random graph, J. London Math. Soc. (2) 48
(1993), 204218.
[K] K. Kunen, Set Theory. An Introduction to Independence Proofs, North Holland, Ams-
terdam (1980).
[MN] H. D. Macpherson and P. M. Neumann, Subgroups of innite symmetric groups, J.
London Math. Soc. (2) 42 (1990), 6484.
[ST1] J. D. Sharp and Simon Thomas, Uniformisation problems and the conality of the
innite symmetric group (to appear in Notre Dame Jouranl of Formal Logic).
[ST2] J. D. Sharp and Simon Thomas, Unbounded families and the conality of the innite
symmetric group (to appear in Arch. Math. Logic).
[Sh-288] S. Shelah, Strong Partition Relations Below the Power Set: Consistency. Was Sier-
pinski Right? II, in Proceedings of the Conference on Set Theory and its Applications
in honor of A. Hajnal and V. T. Sos, Budapest, Sets, Graphs and Numbers, 60 of
Colloquia Mathematica Societatis Janos Bolyai (1991), 637668.
[Sh-400a] S. Shelah, Cardinal arithmetic for skeptics, A.M.S. Bulletin, New Series 26 (1992),
197210.
[Sh-g] S. Shelah, Cardinal Arithmetic, Oxford Logic Guides, 29, Oxford University Press
(1994).
[Sh-430] S. Shelah, Further cardinal arithmetic, Israel J. Math (to appear).
[Sh-513] S. Shelah, PCF and innite free subsets, (in preparation).
MATHEMATICS DEPARTMENT
BILKENT UNIVERSITY
ANKARA
TURKEY
MATHEMATICS DEPARTMENT
THE HEBREW UNIVERSITY
JERUSALEM
ISRAEL
MATHEMATICS DEPARTMENT
RUTGERS UNIVERSITY
NEW BRUNSWICK, NEW JERSEY
USA