7

6
8


r
e
v
i
s
i
o
n
:
2
0
0
4
-
0
4
-
2
2







m
o
d
i
f
i
e
d
:
2
0
0
4
-
0
4
-
2
2


CRITICAL CARDINALITIES AND ADDITIVITY
PROPERTIES OF COMBINATORIAL NOTIONS OF
SMALLNESS
SAHARON SHELAH AND BOAZ TSABAN
Abstract. Motivated by the minimal tower problem, an earlier
work studied diagonalizations of covers where the covers are re-
lated to linear quasiorders (-covers). We deal with two types of
combinatorial questions which arise from this study.
(1) Two new cardinals introduced in the topological study are
expressed in terms of well known cardinals characteristics of
the continuum.
(2) We study the additivity numbers of the combinatorial notions
corresponding to the topological diagonalization notions.
This gives new insights on the structure of the eventual dominance
ordering on the Baire space, the almost inclusion ordering on the
Rothberger space, and the interactions between them.
1. Introduction and overview
We work with two spaces which carry an interesting combinatorial
structure: The Baire space
N
N with eventual dominance

(f

g
if f(n) g(n) for all but nitely many n), and the Rothberger space
P

(N) = A N : A is innite with

(A

B if A B is nite).
We write A

B if A

B and B ,

A.
A subset X of
N
N is unbounded if it is unbounded with respect to

. X is dominating if it is conal in
N
N with respect to

. b is the
minimal size of an unbounded subset of
N
N, and d is the minimal size
of a dominating subset of
N
N.
An innite set A N is a pseudo-intersection of a family T P

(N)
if for each B T, A

B. A family T P

(N) is a tower if it is
linearly quasiordered by

, and it has no pseudo-intersection. t is

1991 Mathematics Subject Classication. 03E17, 06A07, 03E35, 03E10 .
Key words and phrases. -cover, tower, splitting number, additivity number.
The research of the rst author is partially supported by The Israel Science
Foundation founded by the Israel Academy of Sciences and Humanities. Publication
768.
This paper constitutes a part of the second authors doctoral dissertation at
Bar-Ilan University.
1
7
6
8


r
e
v
i
s
i
o
n
:
2
0
0
4
-
0
4
-
2
2







m
o
d
i
f
i
e
d
:
2
0
0
4
-
0
4
-
2
2


2 SAHARON SHELAH AND BOAZ TSABAN
the minimal size of a tower. A family T P

(N) is centered if the

intersection of each (nonempty) nite subfamily of T is innite. p is
the minimal size of a centered family which has no pseudo-intersection.
A family T P

(N) is splitting if for each innite A N there exists

S T which splits A, that is, such that the sets A S and A S are
innite. s is the minimal size of a splitting family.
Let c = 2

0
. The following relations, where an arrow means , are
well-known [3]:
b

1
p t d c

s
No pair of cardinals in this diagram is provably equal, except perhaps p
and t. The Minimal Tower problem, which asks whether it is provable
that p = t, is one of the most important problems in innite combina-
torics, and it goes back to Rothberger (see, e.g., [12]).
New cardinals. In [15], topological notions related to p and t were
compared. In [17] the topological notion related to t (called -covers)
was studied in a wider context. This study led back to several new
combinatorial questions, one of which related to the minimal tower
problem.
Denition 1. For a family T P

(N) and an innite A N, dene

T A = B A : B T. If all sets in T A are innite, we say that
T A is a large restriction of T. Let

be the minimal cardinality of

a centered family T P

(N) such that there exists no innite A N

such that the restriction T A is large and linearly quasiordered by

.
It is not dicult to see that p = min

, t [17]. In Section 2 we
show that in fact, p =

. This existence of a centered family with no

large linearly quasiordered restriction shows that p is combinatorially
larger than asserted in its original denition, and suggests an addi-
tional evidence to the diculty of separating p from the combinatorially
larger cardinal t: Now the consistency of

< t must be established

in order to solve the Minimal Tower problem in the negative.
Denition 2. For functions f, g
N
N, and a binary relation R on N,
dene a subset [f Rg] of N by:
[f Rg] = n : f(n)Rg(n).
7
6
8


r
e
v
i
s
i
o
n
:
2
0
0
4
-
0
4
-
2
2







m
o
d
i
f
i
e
d
:
2
0
0
4
-
0
4
-
2
2


COMBINATORIAL NOTIONS OF SMALLNESS AND ADDITIVITY 3
Next, For functions f, g, h
N
N, and binary relations R, S on N, dene
[h Rg S f] N by:
[f Rg S h] = [f Rg] [g S h] = n : f(n)Rg(n) and g(n)Sh(n).
For a subset X of
N
N and g
N
N, we say that g avoids middles in X
with respect to R, S) if:
(1) for each f X, the set [f Rg] is innite;
(2) for all f, h X at least one of the sets [f Rg S h] and [h Rg S f]
is nite.
X satises the excluded middle property with respect to R, S) if there
exists g
N
N which avoids middles in X with respect to R, S). x
R,S
is the minimal size of a subset X of
N
N which does not satisfy the
excluded middle property with respect to R, S).
The cardinal x = x
<,
was dened in [17]. In section 3 we express all
of the four cardinals x
,
, x
<,
, x
,<
, and x
<,<
in terms of well-known
cardinals. This solves several problems raised in [17].
Additivity of combinatorial notions of smallness. For a nite
subset F of
N
N, dene max(F)
N
N by max(F)(n) = maxf(n) : f
F for each n. A subset Y of
N
N is nitely-dominating if the collection
maxn(Y ) := max(F) : F is a nite subset of Y
is dominating.
We will use the following notations:
B : The collection of all bounded subsets of
N
N,
X : The collection of all subsets of
N
N which satisfy the excluded
middle property with respect to <, ),
D
n
: The collection of all subsets of
N
N which are not nitely dom-
inating; and
D : The collection of all subsets of
N
N which are not dominating.
Thus B X D
n
D. The classes B, X, D
n
, and D are used to
characterize certain topological diagonalization properties [13, 16, 17].
Following [1], we dene the additivity number for classes I J
P(
N
N) with I , J by
add(I, J) = min[F[ : F I and F , J,
and write add(J) = add(J, J). If I contains all singletons, then add(I, J)
non(J), where non(J) = min[J[ : J
N
N and J , J (thus non(B) =
b, non(D) = non(D
n
) = d, and non(X) = x.)
For I, J B, X, D
n
, D, the cardinals add(I, J) bound from be-
low the additivity numbers of the corresponding topological diago-
nalizations. In section 4 we express add(I, J) for almost all I, J
7
6
8


r
e
v
i
s
i
o
n
:
2
0
0
4
-
0
4
-
2
2







m
o
d
i
f
i
e
d
:
2
0
0
4
-
0
4
-
2
2


4 SAHARON SHELAH AND BOAZ TSABAN
B, X, D
n
, D in terms of well known cardinal characteristics of the
continuum. In two cases for which this is not done, we give consistency
results.
2. The cardinal

For our purposes, a lter on a boolean subalgebra B of P(N) is a

family | B which is closed under taking supersets in B and nite
intersections, and does not contain nite sets as elements.
Theorem 3. p =

.
Proof. Let T P

(N) be a centered family of size p which has no

pseudo-intersection. Let B be the boolean subalgebra of P(N) gener-
ated by T. Then [B[ = p. Let | B be a lter of B containing T.
As | does not contain nite sets as elements, | is centered. Moreover,
[|[ = p, and it has no pseudo-intersection.
Towards a contradiction, assume that p <

. Then there exists an

innite A N such that the restriction | A is large, and is linearly
quasiordered by

. Fix any element D

0
A | A. As | A does
not have a pseudo-intersection, there exist:
(1) An element D
1
A | A such that D
1
A

D
0
A; and
(2) An element D
2
A | A such that D
2
A

D
1
A.
Then the sets (D
2
(D
0
D
1
)) A and D
1
A (which are elements of
| A) contain the innite sets (D
0
A)(D
1
A) and (D
1
A)(D
2
A),
respectively, and thus are not

-comparable, a contradiction.
A closely related problem from [17] remains open.
Denition 4. A family Y P

(N) is linearly renable if for each

y Y there exists an innite subset y y such that the family

Y =
y : y Y is linearly

-quasiordered. p

is the minimal size of a

centered family in P

(N) which is not linearly reneable.

Again it is easy to see that p = minp

, t. Thus, a solution of the

following problem may shed more light on the Minimal Tower problem.
Problem 5. Does p = p

?
3. The excluded middle property
Lemma 6. b x
,
x
,<
x
<,
x
<,<
d.
Proof. The inequalities x
,
x
,<
and x
<,
x
<,<
are immediate
from the denitions. We will prove the other inequalities.
7
6
8


r
e
v
i
s
i
o
n
:
2
0
0
4
-
0
4
-
2
2







m
o
d
i
f
i
e
d
:
2
0
0
4
-
0
4
-
2
2


COMBINATORIAL NOTIONS OF SMALLNESS AND ADDITIVITY 5
Assume that Y is a bounded subset of
N
N. Let g
N
N bound Y .
Then g avoids middles in Y with respect to , ). This shows that
b x
,
.
Next, consider a subset Y of
N
N which satises the excluded mid-
dle property with respect to <, <), and let g witness that. Then g
witnesses that Y is not dominating. Thus x
<,<
d.
It remains to show that x
,<
x
<,
. Assume that Y
N
N satises
the excluded middle property with respect to , <), and let g
N
N
avoid middles in Y with respect to , <). Dene g
N
N such that
g(n) = g(n) + 1 for each n. For each f, h Y we have that [f g] =
[f < g], and [f g <h] = [f < g h]. Therefore, g avoids middles in Y
with respect to <, ).
Theorem 7. x
,
= x
,<
= b.
Proof. By Lemma 6, it is enough to show that x
,<
b. Let b

: <
b) be an unbounded subset of
N
N. For each < b dene b
0

, b
1

N
N
by

b
0

(2n) = b

(n)
b
0

(2n + 1) = 0
;

b
1

(2n) = 0
b
1

(2n + 1) = b

(n)
for each n N, and set Y = b
0

, b
1

: < b. Then [Y [ = b. We will

show that Y does not satisfy the excluded middle property with respect
to , <). For each g
N
N, let < b be such that maxg(2n), g(2n +
1) < b

(n) for innitely many n. Then:

[b
0

g <b
1

] = n : b
0

(n) g(n) < b

1

(n)
2n + 1 : 0 g(2n + 1) < b

(n)
is an innite set. Similarly, [b
1

g <b
0

] 2n : 0 g(2n) < b

(n)
is also innite. That is, g does not avoid middles in Y with respect to
, <).
Lemma 8. s x
<,
.
Proof. Assume that Y
N
N is such that [Y [ < s. Let T P(N) be
the family of all sets of the form [f <h], where f, h Y . [T[ < s, thus
there exists an innite subset A of N such that for each X T, either
A X is nite, or A X is nite. As [Y [ < s d, there exists g
N
N
such that for each f Y , g A ,

f A. (In particular, [f <g] is

innite for each f Y .) We may assume that for n , A, g(n) = 0.
Consider any set [f <h] T. If A [f <h] is nite, then the set
[f <g h] n : 0 < g(n), f(n) < h(n)
n A : f(n) < h(n) = A [f <h]
7
6
8


r
e
v
i
s
i
o
n
:
2
0
0
4
-
0
4
-
2
2







m
o
d
i
f
i
e
d
:
2
0
0
4
-
0
4
-
2
2


6 SAHARON SHELAH AND BOAZ TSABAN
is nite. Otherwise, A [f <h] is nite, so we get similarly that
[h <g f] n A : h(n) < f(n)
n A : h(n) f(n) = A [f <h]
is nite. Thus Y satises the excluded middle property with respect
to <, ).
Theorem 9. x
<,
= x
<,<
= maxs, b.
Proof. By Lemmas 6 and 8, we have that maxs, b x
<,
x
<,<
. We
will prove that x
<,<
maxs, b. The argument is an extension of the
proof of Theorem 7.
Let b

be the minimal size of a subset B of

N
N such that B is un-
bounded on each innite subset of N. According to [3], b = b

. Thus
there exists a subset B = b

: < b) of
N
N such that B is increasing
with respect to

and unbounded on each innite subset of N. Let

o = S

: < s) P

(N) be a splitting family. For each < s and

< b dene b
0
,
, b
1
,

N
N by:
b
0
,
(n) =

(n) n S

0 n , S

; b
1
,
(n) =

0 n S

(n) n , S

and set Y = b
i
,
: i < 2, < s, < b. Then [Y [ = 2 s b =
maxs, b. We will show that Y does not satisfy the excluded middle
property with respect to <, <). Assume that g
N
N avoids middles
in Y with respect to <, <). Then the set A = [0 <g] is innite; thus
there exists < s such that the sets A S

and A S

are innite.
Pick < b such that b

A S

g A S

, and > such that

b

A S

g A S

. Then:
[b
0
,
<g <b
1
,
] n A S

: b
0
,
(n) < g(n) < b
1
,
(n)
= n A S

: 0 < g(n) < b

(n)
= n A S

: g(n) < b

(n)
is an innite set. Similarly, the set
[b
1
,
<g <b
0
,
] n A S

: b
1
,
(n) < g(n) < b
0
,
(n)
= n A S

: 0 < g(n) < b

(n)
= n A S

: g(n) < b

(n)
is also innite, because b

; a contradiction.
Remark 10. The cardinal maxs, b is also equal to the nitely split-
ting number fs studied in [8].
7
6
8


r
e
v
i
s
i
o
n
:
2
0
0
4
-
0
4
-
2
2







m
o
d
i
f
i
e
d
:
2
0
0
4
-
0
4
-
2
2


COMBINATORIAL NOTIONS OF SMALLNESS AND ADDITIVITY 7
Several variations of the excluded middle property are studied in the
appendix to the online version of this paper [14].
4. Additivity of combinatorial properties
The additivity number add(I, J) is monotone decreasing in the rst
coordinate and increasing in the second. Our task in this section is
to determine, when possible, the cardinals in the following diagram in
terms of the usual cardinal characteristics b, d, etc. (In this diagram,
an arrow means .)
add(D, D) add(D
n
, D) add(X, D) add(B, D)

add(D
n
, D
n
) add(X, D
n
) add(B, D
n
)

add(X, X) add(B, X)

add(B, B)
4.1. Results in ZFC.
Theorem 11. The following equalities hold:
(1) add(B, D
n
) = add(B, D) = d,
(2) add(D
n
, D
n
) = add(X, X) = add(X, D
n
) = 2; and
(3) add(D, D) = add(B, B) = add(B, X) = b.
Proof. (1) As non(D) = d, it is enough to show that add(B, D
n
) d.
Assume that [I[ < d, and that Y =

iI
Y
i
where each Y
i
is bounded.
For each i I let g
i
bound Y
i
. As [I[ < d, the family maxn(g
i
:
i I) is not dominating; let h be a witness for that. For each nite
F Y , let

I be a nite subset of I such that F

i

I
Y
i
. Then
max(F)

max(g
i
: i

I) ,

h. Thus max(F) ,

h, so Y D
n
.
(2) It is enough to show that add(X, D
n
) = 2. Thus, let
Y
0
= f
N
N : (n)f(2n) = 0 and f(2n + 1) 1
Y
1
= f
N
N : (n)f(2n) 1 and f(2n + 1) = 0
Then the constant function g 1 witnesses that Y
0
, Y
1
X, but Y
0
Y
1
is 2-dominating, and in particular nitely dominating.
(3) It is folklore that add(D, D) = add(B, B) = b see, e.g., [2, full
version] for a proof. It remains to show that add(B, X) b. Let B
be a subset of
N
N which is unbounded on each innite subset of N,
and such that [B[ = b. For each f B let Y
f
= g
N
N : g

f.
(Thus each Y
f
is bounded.) We claim that Y =

fB
Y
f
, X. To
7
6
8


r
e
v
i
s
i
o
n
:
2
0
0
4
-
0
4
-
2
2







m
o
d
i
f
i
e
d
:
2
0
0
4
-
0
4
-
2
2


8 SAHARON SHELAH AND BOAZ TSABAN
this end, consider any function g
N
N which claims to witness that
Y X. In particular, [0 <g] must be innite. Choose f B such that
f [0 <g] ,

g [0 <g], that is, [0 <g <f] is innite. Let A

0
, A
1
be
a partition of [0 <g <f] into two innite sets, and dene f
0
Y
f
by
f
0
(n) = g(n) when n A
0
and 0 otherwise; similarly dene f
1
Y
f
by
f
1
(n) = g(n) when n A
1
and 0 otherwise. Then f
0
, f
1
Y , but both
of the sets [f
0
<g f
1
] and [f
1
<g f
0
] are innite.
4.2. Consistency results. The only cases which we have not solved
yet are add(D
n
, D) and add(X, D). In [2, full version] it was proved
that b add(D
n
, D). In Theorem 2.2 of [10] it is (implicitly) proved
that g add(D
n
, D). Thus
maxb, g add(D
n
, D) add(X, D) d.
Moreover, for any I J, cf(add(I, J)) add(J), and therefore
cf(add(D
n
, D)), cf(add(X, D)) add(D, D) = b.
The notion of ultralter will be used to obtain upper bounds on
add(D
n
, D) and add(X, D). A family | P

(N) is a nonprincipal
ultralter if it is closed under taking supersets and nite intersections,
and cannot be extended, that is, for each innite A N, either A |
or N A |. Consequently, a linear quasiorder
U
can be dened on
N
N by
f
U
g if [f g] |.
The conality of the reduced product
N
N/| is the minimal size of a
subset C of
N
N which is conal in
N
N with respect to
U
.
Theorem 12. For each cardinal number , the following are equiva-
lent:
(1) < add(D
n
, D);
(2) For each -sequence (g

, |

) : < ) with each |

an ultra-
lter on N and each g

N
N there exists g
N
N such that for
each < , [g

g] |

.
Proof. 1 2: For each < let Y

= f
N
N : [f <g

] |

.
Then each Y

D
n
, thus by (1) Y =

<
Y

is not dominating.
Let g
N
N be a witness for that. In particular, for each g , Y

,
that is, [g <g

] , |

. As |

is an ultralter, we have that [g

g] =
N [g <g

] |

.
2 1: Assume that Y =

<
Y

where each Y

D
n
. For each ,
let |

be an ultralter such that Y

/|

is bounded, say by g

N
N [13].
By (2) let g
N
N be such that for each < , [g

g] |

. Then g
witnesses that Y is not dominating: For each f Y , let be such that
7
6
8


r
e
v
i
s
i
o
n
:
2
0
0
4
-
0
4
-
2
2







m
o
d
i
f
i
e
d
:
2
0
0
4
-
0
4
-
2
2


COMBINATORIAL NOTIONS OF SMALLNESS AND ADDITIVITY 9
f Y

. Then [f g

] |

, thus [f <g] [f <g

] [g

g] |

;
therefore [f <g] is innite.
Corollary 13. Assume that | is a nonprincipal ultralter on N. Then
add(D
n
, D) cof(
N
N/|).
Proof. Assume that < add(D
n
, D) and let g

: < ) be any -
sequence of elements of
N
N. For each set |

= |. Then by Theorem
12 there exists g
N
N such that for each , [g

g] |

= |. Thus
g

: < ) is not conal in

N
N/|.
Corollary 14. add(D
n
, D) cf(d).
Proof. Canjar [7] proved that there exists a nonprincipal ultralter |
with cof(
N
N/|) = cf(d). Now use Corollary 13.
Lemma 15. g
N
N avoids middles in Y if, and only if, for each
f Y [f <g] is innite, and the family [f <g] : f Y is linearly
quasiordered by

.
Theorem 16. For any cardinal , the following are equivalent:
(1) < add(X, D);
(2) For each -sequence (g

, T

) : < ), such that each g

N
N,
and for each the restriction T

[0 <g

] is large and linearly

quasiordered by

, there exists h
N
N such that for each <
, the restriction T

[g

h] is large.
Proof. 2 1: Assume that Y =

<
Y

where each Y

X. For
each let g

N
N be a function avoiding middles in Y

, and set
T

= [f <g

] : f Y

. By Lemma 15, T

(N) is linearly
quasiordered by

. As T

[0 <g

] = T

, the restriction is large

and linearly quasiordered by

. By the assumption (2), there exists

h
N
N such that for each < and each f Y

, [f <g

] [g

h] is
innite; therefore h ,

f. Thus h witnesses that Y D.

1 2: Replacing each T

with T

[0 <g

], we may assume that

each A T

is an innite subset of [0 <g

].
For each < let
Y

= f
N
N : [f <g

] T

.
For each A T

and each h
N
N, dene
(1)

h(n) =

(n) 1 n A
maxg

(n), h(n) otherwise

Then [

h <g

] = A, and [

h <h] A. Thus, for each ,

T

= [h <g

] : h Y

(N).
7
6
8


r
e
v
i
s
i
o
n
:
2
0
0
4
-
0
4
-
2
2







m
o
d
i
f
i
e
d
:
2
0
0
4
-
0
4
-
2
2


10 SAHARON SHELAH AND BOAZ TSABAN
As T

is linearly quasiordered by

, we have by Lemma 15 that g

avoids middles in Y

. By (1), Y =

<
Y

is not dominating; let

h
N
N be a witness for that.
For each < and A T

, let

h Y

be the function dened in

Equation 1. Then

h Y , therefore [

h <h] is innite. By the denition

of

h, [

h <h] A [g

h]; therefore the restriction T

[g

h] is
large.
A nonprincipal ultralter | is a simple P

point if it is generated by
a -sequence A

: < ) P

(N) which is decreasing with respect

to

. | is a pseudo-P

point if every family T | with [T[ < has

a pseudo-intersection. Clearly every simple P

point is a pseudo-P

point.
Corollary 17. If | is a simple P

point, then add(X, D) cof(

N
N/|).
Proof. Assume that < add(X, D). Let A

: < ) P

(N) be a
-sequence which generates | and is linearly quasiordered by

, and
set T

= T = A

: < for all < . Assume that g

N
N,
< , are given. We will show that these functions g

are not conal

in
N
N/|.
We may assume that for each < , [0 <g

] = N. Use Theorem 16
to obtain a function h
N
N such that for each < , the restriction
T [g

h] is large. Assume that for some < , [g

h] , |. Then
[h <g

] |, thus there exists < such that A

[h <g

], therefore
A

[g

h] is nite, a contradiction. Thus h + 1 witnesses that the

functions g

are not conal in

N
N/|, therefore < cof(
N
N/|).
In the remaining part of the paper we will consider the remaining
standard cardinal characteristics of the continuum (see [3]). Let u
denote the minimal size of an ultralter base.
Theorem 18. It is consistent (relative to ZFC) that the following
holds:
u = add(D
n
, D) = add(X, D) =
1
<
2
= s = c.
Thus, it is not provable that s add(X, D).
Proof. In [5] a model of set theory is constructed where c =
2
and
there exist a simple P

1
point and a simple P

2
point. The simple P

1
point is generated by
1
many sets, thus u =
1
. As b u, b =
1
as
well.
Nyikos proved that if there exists a pseudo P

point | and > b,

then cof(
N
N/|) = b (see [4]). Thus by Corollary 17, add(X, D) b =

1
in this model. In [4] it is proved that if there exists a pseudo P

point |, then s . Therefore s

2
in this model.
7
6
8


r
e
v
i
s
i
o
n
:
2
0
0
4
-
0
4
-
2
2







m
o
d
i
f
i
e
d
:
2
0
0
4
-
0
4
-
2
2


COMBINATORIAL NOTIONS OF SMALLNESS AND ADDITIVITY 11
Depth
+
(P

(N)) is dened as the minimal cardinal such that there

exists no

-decreasing -sequence in P

(N). (Thus, e.g., t < Depth

+
(P

(N)).)
Each linearly quasiordered family T P

(N) has a conal subfamily

which forms a

-decreasing sequence of length < Depth

+
(P

(N)).
Theorem 19.
(1) If Depth
+
(P

(N)) < d, then add(X, D) = d.

(2) If Depth
+
(P

(N)) = d, then cf(d) add(X, D).

Proof. To prove (1) it is enough to show that for each satisfying
Depth
+
(P

(N)) < d, we have that < add(X, D). To prove (2)

we will show that for each < cf(d), < add(X, D). We will use
Theorem 16, and prove both cases simultaneously.
Assume that Depth
+
(P

(N)) < d (respectively, < cf(d)).

Consider any -sequence (g

, T

) : < ) where each g

N
N,
each T

(N) is linearly quasiordered by

, and the restriction

T

[0 <g

] is large. We must show that there exists h

N
N such
that for each < , the restriction T

[g

<h] is large.
Use the fact that Depth
+
(P

(N)) (respectively, Depth

+
(P

(N)) =
d) to choose for each < a conal subfamily

T

of T

such that
[

T

[ < (respectively, [

T

[ < d).
We may assume that each g

is increasing. For each and each

A T

, let

A
N
N be the increasing enumeration of A. The collection
g

A : < , A T

has less than d many elements and therefore

cannot be dominating. Let h
N
N be a witness for that. Fix < .
For all A T

, there exist innitely many n such that

g

A(n)) = g

A(n) < h(n) h(

A(n)),
that is, A [g

<h] is innite.
Theorem 20. Assume that V is a model of CH and
1
< =

0
.
Let C

be the forcing notion which adjoins many Cohen reals to V .

Then in the Cohen model V
C
, the following holds:
add(D
n
, D) = s = a = non(/) =
1
< cov(/) = add(X, D) = c.
Proof. The assertions s = a = non(/) =
1
< cov(/) = c are
well-known to hold in V
C
, see [3]. It was proved by Kunen [9] that
V
C
[= Depth
+
(P

(N)) =
2
. As cov(/) d, we have that d = c =
in this model. If =
2
, use Theorem 19(1) and the fact that d is
regular in this model to obtain d add(X, D). Otherwise use Theorem
19(2) and the fact that Depth
+
(P

(N)) =
2
< = d to obtain this.
In [6, 11] it is proved that there exists a nonprincipal ultralter |
in V
C
such that cof(
N
N/|) =
1
. By Corollary 13, we have that
add(D
n
, D) =
1
in V
C
.
7
6
8


r
e
v
i
s
i
o
n
:
2
0
0
4
-
0
4
-
2
2







m
o
d
i
f
i
e
d
:
2
0
0
4
-
0
4
-
2
2


12 SAHARON SHELAH AND BOAZ TSABAN
In particular, the cardinals add(D
n
, D) and add(X, D) are not prov-
ably equal.
Corollary 21. It is not provable that add(X, D) cf(d).
Proof. Use Theorem 20 with =

1
. In V
C
, d = c =

1
, therefore
cf(d) =
1
< add(X, D) in this model.
Remark 22. In the remaining canonical models of set theory which are
used to distinguish between the various cardinal characteristics of the
continuum (see [3]), maxb, g = d holds, and therefore add(D
n
, D) =
add(X, D) = d too. These models show that none of the follow-
ing is provable: mincov(^), r add(X, D) (Random reals model),
add(D
n
, D) maxcov(^), s (Hechler reals model), add(D
n
, D)
maxnon(^), cov(^) (Laver reals model), and add(D
n
, D) maxu,
a, non(^), non(/) (Miller reals model).
Collecting all of the consistency results, we get that the only possible
additional lower bounds on add(X, D) are cov(/) and e (observe that
e cov(/) [3].)
Problem 23. Is cov(/) add(X, D)? And if not, is e add(X, D)?
No additional cardinal characteristic can serve as an upper bound
on add(D
n
, D).
Another question of interest is whether add(D
n
, D) or add(X, D)
appear in the lattice generated by the cardinal characteristics with
the operations of maximum and minimum. In particular, we have the
following.
Problem 24. Is it provable that add(D
n
, D) = maxb, g?
We have an indication that the answer to Problem 24 is negative,
but this is a delicate matter which will be treated in a future work.
References
[1] T. Bartoszy nski and H. Judah, Set Theory: On the structure of the real line,
A. K. Peters, Massachusetts: 1995.
[2] T. Bartoszynski, S. Shelah, and B. Tsaban, Additivity properties of topological
diagonalizations, The Journal of Symbolic Logic 68 (2003), 12541260. (Full
version: http://arxiv.org/abs/math.LO/0112262)
[3] A. R. Blass, Combinatorial cardinal characteristics of the continuum, to appear
in: Handbook of Set Theory (eds. M. Foreman, et. al.).
[4] A. R. Blass and H. Mildenberger, On the conality of ultrapowers, Journal of
Symbolic Logic 64 (1999), 727736.
[5] A. R. Blass and S. Shelah, There may be simple P
1
- and P
2
-points, and the
Rudin-Keisler ordering may be downward directed, Annals of Pure and Applied
Logic 33 (1987), 213243.
7
6
8


r
e
v
i
s
i
o
n
:
2
0
0
4
-
0
4
-
2
2







m
o
d
i
f
i
e
d
:
2
0
0
4
-
0
4
-
2
2


COMBINATORIAL NOTIONS OF SMALLNESS AND ADDITIVITY 13
[6] R. M. Canjar, Countable ultraproducts without CH, Annals of Pure and Applied
Logic 37 (1988), 179.
[7] R. M. Canjar, Conalities of countable ultraproducts: the existence theorem,
Notre Dame J. Formal Logic 30 (1989), 539542.
[8] A. Kamburelis and B. W eglorz, Splittings, Archive for Mathematical Logic 35
(1996), 263277.
[9] K. Kunen, Inaccessibility Properties of Cardinals, Doctoral Dissertation, Stan-
ford, 1968.
[10] H. Mildenberger, Groupwise dense families, Archive for Mathematical Logic
40 (2001), 93112.
[11] J. Roitman, Non-isomorphic H-elds from non-isomorphic ultrapowers, Math.
Z. 181 (1982), 9396.
[12] F. Rothberger, On some problems of Hausdor and of Sierpi nski, Fund. Math.
35 (1948), 2946.
[13] M. Scheepers and B. Tsaban, The combinatorics of Borel covers, Topology and
its Applications 121 (2002), 357382.
http://arxiv.org/abs/math.GN/0302322
[14] S. Shelah and B. Tsaban, Critical cardinalities and additivity
properties of combinatorial notions of smallness (online version),
http://arxiv.org/abs/math.LO/0304019
[15] B. Tsaban, A topological interpretation of t, Real Analysis Exchange 25
(1999/2000), 391404.
http://arxiv.org/abs/math.LO/9705209
[16] B. Tsaban, A diagonalization property between Hurewicz and Menger, Real
Analysis Exchange 27 (2001/2002), 757763.
http://arxiv.org/abs/math.GN/0106085
[17] B. Tsaban, Selection principles and the minimal tower problem, Note di
Matematica, to appear.
http://arxiv.org/abs/math.LO/0105045
Appendix A. Variations of the excluded middle property
Denition 25. For a subset X of
N
N, g
N
N, and R, S , <, we
say that g quasi avoids middles in X with respect to R, S) if:
(1) g is unbounded;
(2) for all f, h X at least one of the sets [f Rg S h] and [h Rg S f]
is nite.
A function g
N
N satisfying item (2) above is said to weakly avoid
middles in X with respect to R, S). X satises the quasi excluded
middle property (respectively, weak excluded middle property) with re-
spect to R, S) if there exists g
N
N which quasi (respectively, weakly)
avoids middles in X with respect to R, S). x

R,S
(respectively, x

R,S
) is
the minimal size of a subset X of
N
N which does not satisfy the quasi
(respectively, weak) excluded middle property with respect to R, S).
Lemma 26. The following inequalities hold:
7
6
8


r
e
v
i
s
i
o
n
:
2
0
0
4
-
0
4
-
2
2







m
o
d
i
f
i
e
d
:
2
0
0
4
-
0
4
-
2
2


14 SAHARON SHELAH AND BOAZ TSABAN
(1) x

,
x

,<
x

<,
x

<,<
,
(2) x

,
x

,<
x

<,
x

<,<
;
(3) For each R, S , <, x
R,S
x

R,S
x

R,S
.
Proof. (1) and (2) are proved as in Lemma 6. We will prove the rst
inequality of (3), the other one being immediate from the denitions.
Assume that Y
N
N satises [Y [ < x
R,S
. Let i
N
N be the identity
function. Set Y

= Y i. Then [Y

[ < x
R,S
, thus there exists g
N
N
which avoids middles in Y

. In particular, the set [i Rg] is innite, thus

g is unbounded, so g quasi avoids middles in Y .
Theorem 27. The following assertions hold:
(1) Every subset X of
N
N satises the weak excluded middle prop-
erty with respect to <, ). Thus, x

<,
= x

<,<
= .
(2) x

,
= x

,<
= x

,
= x

,<
= b;
(3) x

<,
= x

<,<
= maxs, b.
Proof. (1) Let o
N
N be the constant zero function. Then for each
f, h
N
N, the set [f <o h] is nite.
(2) By Lemmas 6 and 26, it is enough to show that x

,<
b. But
this is, actually, what is shown in the proof of Theorem 7.
(3) By Lemmas 6, 8 and 26, it is enough to show that x

<,<

maxs, b. The proof is identical to the proof of Theorem 9, since for
an unbounded g
N
N, the set [0 < g] is innite, as required there.
Institute of Mathematics, Hebrew University of Jerusalem, Givat
Ram, 91904 Jerusalem, Israel, and Mathematics Department, Rutgers
University, New Brunswick, NJ 08903, U.S.A.
E-mail address: shelah@math.huji.ac.il
Department of Mathematics and Computer Science, Bar-Ilan Uni-
versity, Ramat-Gan 52900, Israel
E-mail address: tsaban@macs.biu.ac.il