You are on page 1of 8

3

8
5


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
3
-
3
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
3
-
3
1


ON A CONJECTURE OF TARSKI ON PRODUCTS OF CARDINALS
Thomas Jech
1
and Saharon Shelah
2
Abstract
3
We look at an old conjecture of A. Tarski on cardinal arithmetic and show that if a
counterexample exists, then there exists one of length
1
+.
1
Supported partially by an NSF grant. I wish to express my gratitude to the Mathematical Institute
of the Eidgenossische Technische Hochschule in Z urich for its hospitality during my visit.
2
Publication #385. Supported partially by the B.S.F.
3
AMS classication: 03E

Keywords: Cardinal arithmetic, singular cardinals problem, pcf


Typeset by A
M
S-T
E
X
1
3
8
5


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
3
-
3
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
3
-
3
1


2 THOMAS JECH
1
AND SAHARON SHELAH
2
In the early days of set theory, Hausdor and Tarski established basic rules for ex-
ponentiation of cardinal numbers. In [T] Tarski showed that for every limit ordinal ,

<

||
, and conjectured that
(1)

<

||
holds for every ordinal and every increasing sequence {

}
<
such that lim
<

= .
He remarked that (1) holds for every countable ordinal .
Remarks. 1. The left hand side of (1) is less than or equal to the right hand side.
2. If has || disjoint conal subsets then the equality (1) holds. Thus the rst limit
ordinal that can be the length of a counterexample to (1) is
1
+.
[Proof. Let {A
i
: i < ||} be disjoint conal subsets of . Then

<

i<||

A
i

i<||

||
.]
It is not dicult to see that if one assumes the Singular Cardinals Hypothesis then (1)
holds. With the hindsight given by results obtained in the last twenty years, it is also not
dicult to nd a counterexample to Tarskis conjecture. For instance, using the model
described in [M], one can have an increasing sequence of cardinals of length =
1
+
whose product does not satisfy (1). The purpose of this note is to show that if Tarskis
conjecture fails then it fails in this specic way. Namely, if there is a counterexample then
there is one of length
1
+.
The main result of this paper is the following:
Theorem. A necessary and sucient condition for Tarskis conjecture to fail is the exis-
tence of a singular cardinal

of conality
1
such that

>

1
and

1
>
+

0
.
If

is a cardinal that satises the condition then the sequence {

}
<
1
{
+n
}
n<
is a counterexample to (1):

<
1

n<

+n
=

1

+

0
<
+
|
1
+|
.
Such a cardinal exists in one of Magidors models, e.g. when

1
+
1
is a strong limit,

1
+
1

1
=

1
+
1
++2
and

1
+
1
+

0
=

1
+
1
++1
.
Also, if >

1
is a strong limit singular cardinal of conality
1
such that

1
>

+(2

0
)
+
then we have a counterexample as (
+
)

0
<
+(2

0
)
+
(by [ShA2, Ch. XIII,
5.1]).
The rest of this paper is devoted to the proof that the condition is necessary.
Assume that Tarskis conjecture fails, and let be a limit ordinal for which there exists
a sequence {

}
<
that gives a counterexample:
(2)

<

<

,
where
= || and = lim
<

.
3
8
5


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
3
-
3
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
3
-
3
1


ON A CONJECTURE OF TARSKI ON PRODUCTS OF CARDINALS 3
Lemma 1. If (2) holds then cf < < , and there exists an ordinal < such that

>

.
Proof. If (2) holds then does not have || disjoint conal subsets, and it follows that
is not a cardinal, and that cf < ||.
Assuming that

holds for all < , we pick a conal sequence {


i
}
i<cf
with
limit , and then

= (

i<cf

i
)

i<cf

i<cf

cf
=

i<cf

<

,
contrary to (2).
Now consider the shortest counterexample to Tarskis conjecture.
Lemma 2. If is the least ordinal for which (2) holds then = + where is an
uncountable cardinal.
Proof. Without loss of generality, the sequence is continuous. (We can replace each

by the limit of the sequence at , for each limit ordinal .)


Let = ||. We claim that for every limit ordinal < ,

<

. If this were not


true then, because > , there would be a limit ordinal such that < and that

>

<

, which would make the sequence {

}
<
a counterexample to
Tarskis conjecture as well, contrary to the minimality of .
Thus = + for some limit ordinal . It is clear that the sequence
{

: or > }
of length + is also a counterexample, and by the minimality of we have = +.
Now consider the least ordinal such that

>

. We shall show that cf = (and


so is a regular uncountable cardinal). We also establish other properties of

.
Lemma 3. If Tarskis conjecture fails, then there is a cardinal

of uncountable conality
such that > , and that
for every < ,

<

(3)

>
+

0
. (4)
Proof. Let = + be the least ordinal for which (2) holds, for some increasing continuous
sequence {

: < } with limit , and let be the least ordinal such that

>

.
First we observe that for every < ,

<

. This is because if

then

>

, contradicting the minimality of .


As a consequence, we have cf : otherwise, we would have

<

<

, a contradiction. Also, if = lim


icf

i
, then

=
_

i<cf

i
_

i<cf

i<cf

cf
and so we have

cf
=

.
3
8
5


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
3
-
3
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
3
-
3
1


4 THOMAS JECH
1
AND SAHARON SHELAH
2
Since

<

, we have

cf

cf
, and so

cf
=

, and

cf
>

<

. Hence the sequence


{

: cf or > }
of length cf + is also a counterexample, and it follows that = cf.
For every limit < we have

<

, and in particular

<

. Since

>

,
we have > . Finally,

<

=

<

n<

+n
=

0
=

0
,
and because

>

<

, we have

>

0
. Since = lim
+n
n
+ ,
we have

>
+

0
,
completing the proof.
The cardinal

obtained in Lemma 3 satises all the conditions stated in the Theorem


except for the requirement that its conality be
1
. Thus the following lemma will complete
the proof:
Lemma 4. Let

be a singular cardinal of conality >


1
such that > and that
(5) for every < ,

<

.
Assume further that for every ,
1
< < , of conality
1
,
(6) if for every < ,

1
<

, then

1

+

0
.
Then

0
.
Lemma 4 implies that the least in Lemma 3 has conality
1
, and the theorem follows.
The rest of the paper is devoted to the proof of Lemma 4. We use the second authors
analysis of pcf.
Denition. If A is a set of regular cardinals, let
A = {f : domf = A and f() < for all A}.
If I is an ideal on A then A/I is a partially ordered set under
f
I
g i { : f() > g()} I,
and similarly for lters on A. If D is an ultralter on A, then A/D is a linearly ordered
set, and cf(A/D) denotes its conality. Let
pcf(A) = {cf(A/D) : D an ultralter on A}.
It is clear that
A pcf(A), A
1
A
2
implies pcf(A
1
) pcf(A
2
), and
pcf(A
1
A
2
) = pcf(A
1
) pcf(A
2
),
and it is not dicult to show (using ultrapowers of ultrapowers) that
if |pcf(A)| < min A then pcf(pcf(A)) = pcf(A) and
pcf(A) has a greatest element.
3
8
5


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
3
-
3
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
3
-
3
1


ON A CONJECTURE OF TARSKI ON PRODUCTS OF CARDINALS 5
Theorem (Shelah [Sh345]). If 2
|A|
< min(A) then there exists a family {B

: pcf(A)}
of subsets of A such that
(7) for every ultralter D on A, cf(A/D) = the least such that B

D.
For every pcf(A) there exists a family {f

: < } A such that


(8)
< implies f

< f

mod J
<
, where J
<
is the ideal generated
by {B

: < }, and the f

s are conal in B

mod J
<
.

An immediate consequence of (7) is that |pcf(A)| 2


|A|
. The sets B

( pcf(A)) are
called generators for A. Note that max B

= when A, and that max(pcf(B

)) =
for all .
We shall use some properties of generators.
Lemma 5 [Sh345]. Let B

be generators for A. For every X A there exists a nite set


F pcf(X) such that X

{B

: F}.
Proof. Let Y = pcf(X), and assume that the lemma fails. Then {X B

: Y } has
the nite intersection property and so there is an ultralter D on A such that X D and
B

/ D for all Y . Let = cf(A/D). Then pcf(X) and by (7), B

D. A
contradiction.
For each X A, let s(X) (a support of X) denote a nite set F pcf(X) with the
property that X

F
B

.
The set pcf(A) has a set of generators that satisfy a transitivity condition:
Lemma 6 [Sh345]. Assume that 2
|A|
< min(A) and let

A = pcf(A). Then pcf(

A) =

A
and

A has a set of generators {B

:

A} that satisfy, in addition to (7),
(9) if B

then B

We use the transitivity to prove the next lemma.


Lemma 7. Assume that 2
|A|
< min(A), let

A = pcf(A), let B

,

A, be transitive
generators for

A, and for each X

A let s(X) be a support of X. If A =

iI
A
i
, then

A =
_
_
pcf(B

) : pcf
_
_
iI
s(pcf(A
i
))
__
.
3
8
5


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
3
-
3
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
3
-
3
1


6 THOMAS JECH
1
AND SAHARON SHELAH
2
Corollary. max(

A) = max pcf

iI
s(pcf(A
i
)).
[Proof of Corollary. Let = max(

A); pcf(B

) for some in pcf(

i
s(A
i
)). Since
max(pcf(B

)) = , we have .]
Proof. Let X =

iI
s(pcf(A
i
)) and F = s(X). We have
A =
_
iI
A
i

_
iI
pcf(A
i
)
_
iI
_
{B

: s(pcf(A
i
))} =
=
_
{B

: X}
_
{B

:
_
F
B

}
_
F
B

(the last inclusion is a consequence of transitivity (9)). Therefore

A = pcf(A) pcf(
_
F
B

) =
_
F
pcf(B

)
_
{pcf(B

) : pcf(X)}.

Toward the proof of Lemma 4, let {


i
: i < } be a continuous increasing sequence of
limit ordinals of conality < , such that lim
i

i
= , 2

<

0
, and that for all i < ,
(10) for all <
i
,

<

i
.
Lemma 8. There is a closed unbounded set C such that for all n = 1, 2, . . . ,
(11) max pcf({

i
+n
: i C})
+n
.
Proof. We show that for each n there exists a closed unbounded set C
n
such that
max pcf({

i
+n
: i C
n
})
+n
. To prove this, let n 1 be xed and let A = {

i
+n
:
i < }. Let be the least element of pcf(A) above
+n
(if there is none there is nothing to
prove). Let {B

: pcf(A)} be subsets of A that satisfy (7), and let {S

: pcf(A)}
be the subsets of such that B

= {

i
+n
: i S

}. It suces to prove that the set


S

+1
S

+n
contains a closed unbounded set.
Thus assume that the set S = (S

+1
S

+n
) is stationary. Let J
<
be the ideal
on A generated by {B

: < }. By Shelahs Theorem there exists a family {f

: < }
in A such that < implies f

< f

mod J
<
. Since all the sets B

, <

, are
bounded, we get a family {g

: < } of functions on S such that g

(i) <

i
+n
for
all i S, and such that < implies that g

(i) < g

(i) for eventually all i S. This


contradicts the results in [GH] by which, under the assumption (5), any family of almost
disjoint functions in

iS

i
+n
has size at most
+n
.
Proof of Lemma 4. Let be a singular cardinal of conality >
1
that satises (5) and
(6). Let be a regular cardinal such that

<

. We shall prove that


+

0
.
Let {
i
: i < } be an increasing continuous sequence that satises (10), and let C be
a closed unbounded subset of given by Lemma 8. Let
S = {i C : cf
i
=
1
}.
As
2
, S is a stationary subset of .
3
8
5


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
3
-
3
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
3
-
3
1


ON A CONJECTURE OF TARSKI ON PRODUCTS OF CARDINALS 7
Lemma 9. There exist regular cardinals
i
, i S, such that for each i S,

i
<
i

1
, and an ultralter D on S such that cf(

iS

i
/D) = .
Proof. Let I
0
be the nonstationary ideal on S. There are conal subsets X of

of
size |X| = . For every such set X, let F
X


iC
[

i
]

be the function dened by


F(i) = X

i
. Then when X = Y , F
X
and F
Y
are eventually distinct.
For every i S we have

1
(by (10)), and so there exist I
0
-distinct functions
in

iS

1
. [f and g are I
0
-distinct if {i : f(i) = g(i)} I
0
.]
Consider the partial ordering f <
I
0
g dened by {i : f(i) g(i)} I
0
; since I
0
is
-complete, <
I
0
is well-founded. Let g be a <
I
0
-minimal function with the property that
g(i)

1
and that there are there are I
0
-distinct functions below g.
Let I be the extension of I
0
generated by all the stationary subsets X of S that have the
property that g is not minimal on I
0
[X] (i.e. there is a function g

such that g

(i) < g(i)


almost everywhere on X and below g

there are I
0
-distinct functions).
Claim. I is a normal -complete ideal on S.
[Proof. Let X
i
, i < , be sets in I, and let for each i < , g
i
< g on X
i
and h
i

: <
witness that X
i
I. Then one constructs witnesses g and

: < for X = {j :
j

i<j
X
i
} by letting g(j) = g
i
(j) and

h

(j) = h
i

(j) where i is some i < j such that


j X
i
.
For example, let us show that

h

and

h

are I
0
-distinct if = . Assume that

h

=

h

on a stationary subset S
1
of S. Then on a stationary subset S
2
of S
1
the i less than j S
2
chosen such that j X
i
is the same i, and we have h
i

= h
i

on S
2
, a contradiction.]
Let {h

: < } be a family of I
0
-distinct functions below g.
Claim. For every h <
I
g there is some
0
< such that for all
0
, h <
I
h

.
[Proof. If there are many s such that h h

on an I-positive set, then (because 2

< )
there is an I-positive set X such that h h

on X for many , but this contradicts the


denition of I.]
Using this Claim, one can construct a <
I
-increasing -sequence (a subsequence of {h

:
< }) of functions that is <
I
-conal in

iS
g(i). Let
i
= cfg(i), for each i S.
The product

iS

i
has a <
I
-conal <
I
-increasing sequence of length , and since I is
a normal ideal, we have
i
>

i
for I-almost all i. Now if D is any ultralter extending
the dual of I, D satises cf(

iS

i
/D) = .
Back to the proof of Lemma 4. For each i S we have a regular cardinal
i
such that

i
<
i

1
. By the assumption (6) we have

i
+

0
, and so
i

i
+

0
.
We use the following result:
Theorem (Shelah [ShA2], Chapter XIII, 5.1). Let

be such that

0
<
+
. Then
for every regular cardinal such that

<
+

0
there is an ultralter U on such
that cf(

+n
/U) = .
We apply the theorem to each

i
, and obtain for each i S an ultralter U
i
on such
that cf(

i
+n
/U
i
) =
i
. Combining the ultralters U
i
with the ultralter D on S
3
8
5


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
3
-
3
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
3
-
3
1


8 THOMAS JECH
1
AND SAHARON SHELAH
2
from Lemma 9 we get an ultralter U on the set
A = {

i
+n
: i S, n = 1, 2, . . . }
such that cf(A/U) = . Hence pcf(A).
We shall now complete the proof of Lemma 4 by showing that max pcf(A)
+

0
.
We have A =

n=1
A
n
, where
A
n
= {

i
+n
: i S},
and since 2
|A|
= 2

< min(A), we apply the corollary of Lemma 7 and get


max pcf(A) = max pcf

_
n=1
s(pcf(A
n
)),
where for each n, s(pcf(A
n
)) is a nite subset of pcf(pcf(A
n
)) = pcf(A
n
).
Let E =

n=1
s(pcf(A
n
)). Since (by Lemma 8) max pcf(A
n
)
+n
for each n, E is
a countable subset of
+
. Hence max pcf(E)
+

0
, and so
max pcf(A) = max pcf(E)
+

0
.

References
[[GH]] F. Galvin and A. Hajnal, Inequalities for cardinal powers, Annals of Math. 101 (1975),
491498.
[[M]] M. Magidor, On the singular cardinals problem I, Israel J. Math. 28 (1977), 131.
[[ShA2]] S. Shelah, Proper Forcing, SpringerVerlag Lecture Notes 940, 1982.
[[Sh345]] , Products of regular cardinals and cardinals invariants of products of Boolean
algebras, Israel J. Math. 70 (1990), 129187.
[[Sh355]] ,
+1
has a Jonsson algebra, preprint.
[[T]] A. Tarski, Quelques theor`emes sur les alephs, Fundamenta Mathematicae 7 (1925),
114.
T. Jech, Department of Mathematics, The Pennsylvania State University, University
Park, PA 16802
S. Shelah, Department of Mathematics, The Hebrew University, Jerusalem, Israel

You might also like